AI Field: Recent Technological Highlights and Investment Opportunities

Image Source: Generated by Wujie AI

Editor's note: The development of artificial intelligence technology is advancing rapidly, and many AI startups are standing out in the field. In this article, let's take a look at which promising AI startups have been chosen by investors such as Sequoia Capital and Kleiner Perkins. This article is compiled and we hope it will inspire you.

If you only have a few minutes of free time, here are the most exciting AI startups that investors, operators, and founders should know about.

Advancing human health. Startups are using AI technology to improve medical outcomes and design new treatment methods. For example, Alife uses AI to improve in vitro fertilization treatment, providing better chances of conception for patients. Over time, the company's methods may fundamentally disrupt the cost structure of the industry. NewLimit is another medical startup using AI technology, seeking better methods to treat previously incurable stubborn diseases.
Serving enterprises with AI. Most products of generative AI serve consumers. However, a few promising companies directly meet the needs of enterprises, building products containing internal data based on the company's guidelines. Glean, Lamini, Dust, and Lance represent this trend.
Limiting AI with AI. The AI revolution may bring many new opportunities, but it also poses many threats. In particular, generative AI makes creating realistic written information easy, increasing the quantity and complexity of "phishing" scams aimed at obtaining personal information from recipients. Companies like Abnormal Security can use AI to detect malicious AI information and prevent such attacks.
AI startups are global. Despite the presence of industry giants like OpenAI and Google in the United States, promising startups are emerging globally. Mistral, based in Paris, is building open-source large-scale language models, with one founder believing they can rival OpenAI. German company Sereact has also developed impressive AI robot products and signed contracts with industrial giants.

AI is the main theme of this year's tech story. Since the last "What to Watch in AI" series report, the field has continued to attract capital, talent, and attention. Of course, not all attention is positive. While people are generally excited about the technology's capabilities, in the past four months, industry heavyweights have expressed their concerns, and regulatory agencies have begun designing some preventive measures. In the coming months and years, AI will have a comprehensive impact on our lives and create new winners and losers globally.

Our "What to Watch" series aims to help readers prepare for the upcoming era and foresee the future more clearly. For those who want to understand the emerging technology at the forefront of AI and leverage the ongoing changes, this is a great starting point. To this end, we have invited the most impressive investors and founders in the AI field to introduce the startups they believe have the most potential.

1. Alife

Using AI to improve in vitro fertilization technology

In any reproductive process, there are moments that require human decision-making, with "ovarian stimulation" and "embryo selection" being the two most relevant aspects to IVF.

"Ovarian stimulation" involves determining the dosage of drugs a patient receives to stimulate the growth of follicles in the ovaries and when to administer the trigger injection to induce follicles to release eggs. The timing of the trigger injection is crucial; if it's too early, immature eggs may be obtained, and if it's too late, overly mature eggs may be obtained, or not as many eggs as possible.

"Embryo selection" refers to choosing which fertilized egg to use and implant. Currently, clinical doctors and embryologists, like most medical professionals, rely on their own experience and training, morphological grading systems, and repeated trials to make decisions. If the dosage or timing is not right in one cycle, adjustments are made in the next cycle. This high demand for professional competence among doctors varies, and their techniques are crucial to the results. For the severely restricted market of reproduction, this means high costs, especially if you want to see the best results.

Alife is building AI tools to improve in vitro fertilization (IVF) outcomes. The company uses AI tools to provide "superpowers" to practitioners by leveraging large input and outcome datasets, enhancing their decision-making accuracy. Now, through a simple interface, doctors can input patient characteristics and receive precise recommendations at key moments in the reproductive process, based on results from thousands of previous cycles. These datasets come from existing patient information, and as each patient uses Alife's product, these datasets will improve.

These tools will change the nature of the reproductive industry. Alife's research shows that their machine learning models can help doctors optimize 50% of trigger timings and help retrieve an average of three mature eggs, two fertilized eggs, and one embryo. Alife's product can significantly broaden the channels for infertility treatment, reduce the cost for each patient by lowering the required drug dosage and increasing the success rate of IVF cycles. This will also make the competitive environment for doctors more equitable, allowing those lacking firsthand experience to gain broader knowledge and information.

Ultimately, you can imagine that Alife's tools can provide all the information for a judgment moment in a process and allow non-doctors to operate, significantly changing the cost structure and availability of the industry. More importantly, data-driven precision medicine will enhance (or ultimately replace) a person's judgment through personalized recommendations, which is not unique to the IVF field. In the entire medical field, there are thousands of such moments, and we have the opportunity to use data to greatly change the results and accessibility of key procedures and treatments.

Rebecca Kaden, General Partner, Union Square Ventures

2. Glean

Enterprise search

In the workplace, finding the required information accurately should be quick and simple when needed. However, this is not always the case due to the use of numerous applications and the generation of a large amount of data and documents by everyone in their work. With the exponential growth of "knowledge" and the increasing dispersion of work nature, finding existing knowledge takes longer. In other words, "searching for something" at work is quite challenging.

To help employers address this issue, Arvind Jain and his team built an AI-based workplace unified search platform, Glean. It equips employees with an intuitive work assistant to help them accurately find the required information and proactively discover what they should know.

The company's mission was simple from the start: to help people find answers to all workplace questions faster, reducing frustration and time wastage. However, the company's achievements have far exceeded the scope of search. For example, Glean can not only search all workplace applications and knowledge bases (Slack, Teams, Google Drive, Figma, Dropbox, Coda, etc.) but also understand natural language and context, personalize user interactions based on people's roles and internal/external relationships. It intelligently displays the most popular and validated information within the company, helping you discover what the team knows and maintain consistency, all in a permissioned manner.

As organizations become increasingly dispersed and knowledge becomes more fragmented, intuitive work assistants like Glean are no longer just a bonus but a key tool to improve employee productivity. The company's development will break down barriers to progress, creating a more positive and productive work experience.

In addition, Glean's search technology enables it to bring generative AI into the workplace while complying with strict enterprise permissions and data management requirements. Today, one of the main obstacles preventing enterprises from delivering AI applications at scale is their inability to implement proper governance controls. By inserting real-time data permissions into the enterprise's internal environment, Glean has become an ideal solution to help enterprises address governance issues at scale and allow them to confidently use their internal data for model training and inference, thus leveraging enterprise AI data platforms/vector storage.

Josh Coyne, Partner, Kleiner Perkins

3. Lance

Storage and management of multimodal data

We have all played with Midjourney, and most of us have seen the demo of GPT-4. Midjourney (text to image) and GPT-4 (image to text/code) illustrate the potential of bridging different forms of media such as text, images, and audio when models become multimodal. While the current AI hype mostly revolves around text-based models, multimodal models are the key to building more accurate world representations.

As we usher in the next wave of AI applications in industries such as robotics, healthcare, manufacturing, entertainment, and advertising, more and more companies will build on multimodal models. Companies like Runway and Flair.ai are exemplary leaders in their respective fields, with their products meeting significant user demand, and existing companies like Google have also begun releasing similar multimodal features.

However, using multimodal models presents a challenge: how to store and manage data? Traditional storage formats like Parquet are not optimized for unstructured data, leading to performance issues for large language model teams during data loading, analysis, evaluation, and debugging. Additionally, due to the lack of a single source of truth, large language model workflows are more prone to errors in some subtle aspects. Lance is a company that has recently emerged to address this challenge. Companies like Midjourney and WeRide are converting PB-level datasets to Lance format, which has shown significant performance improvements compared to traditional formats like Parquet and TFRecords, and also reduces incremental storage costs by an order of magnitude.

Lance does not stop at storage; they have recognized the need to rebuild the entire data management stack to better adapt to the world we are moving towards, where unstructured, multimodal data will become enterprises' most valuable assets. Their first platform product, LanceDB (currently in beta), provides a seamless embedded experience for developers looking to build multimodal capabilities into their applications.

Lance is just one example of a company leading developers into a multimodal future, and I am very excited to see the emergence of other technologies driving the development of multimodal applications. With the advancement of artificial intelligence, such a future will soon become a reality. — Saar Gur, General Partner, CRV

4. Abnormal Security

Curbing the Wave of AI-Enhanced Cyber Attacks

I am an unabashed optimist when it comes to generative AI, but I am not naive about this issue. For example, I am concerned about the sharp increase in "social engineering" attacks (such as spear-phishing) that often use emails to extract sensitive information, especially since the rise of ChatGPT last year.

According to data from Abnormal Security, the number of attacks per thousand people has skyrocketed from less than 500 to over 2500 in the past year. The complexity of the attacks has also increased sharply. Just as any student can use ChatGPT to write a perfect essay, ChatGPT can also be used to send grammatically perfect, personalized fraudulent messages.

According to the FBI, targeted "business email compromise" attacks have caused over $50 billion in losses since 2013. And the situation is only getting worse. Countless cybercriminals and other bad actors use black hat tools like "WormGPT" (a chatbot designed to mine malicious data to orchestrate the most persuasive, large-scale fraud activities) for fraudulent activities every day.

Fortunately, Abnormal's co-founders Evan Reiser and Sanjay Jeyakumar are working hard to use AI to combat this threat. You can think of this as using AI to fight AI. Historically, email security systems scan for known bad behaviors, such as specific IP addresses or attempts to access personally identifiable information (PII).

With the power of AI, Abnormal disrupts all of this. With AI's support, many attacks appear legitimate, so Abnormal's approach is to fully understand known good behaviors, even the slightest deviations. The company uses large language models to build detailed representations of its digital internal and external operations, such as who typically interacts with each other and what they might interact about. If my partner Reid Hoffman sends me an email saying, "Hey, please send me the latest materials from Inflection.AI," Abnormal's AI engine quickly realizes that Reid rarely starts with "Hey," rarely sends single-sentence messages, and has never asked me to send him files about Inflection.AI. (As a co-founder and board member of the company, he has more access to these files than I do!)

Not surprisingly, as concerns about the security of generative AI continue to grow, the demand from Abnormal's enterprise clients is also accelerating. I find Abnormal's success very reassuring, as it is able to quickly leverage AI to address the problems accelerated by AI during this period of disruptive technological change. Bad actors often enjoy a long first-mover advantage during such periods, as they can innovate without worrying about product quality, security, or regulatory agencies that have not yet formulated new laws.

At the same time, it is understandable that tech startups are focused on developing powerful new use cases for their innovations rather than stopping illegal or destructive innovation. But as with all issues related to AI, the potential for network disruption from AI misuse is staggering. Thanks to the foresight of the Abnormal team, the new normal of cybercrime may at least become less easy to achieve. — Sam Motamedi, Partner, Greylock

5. Dust

Enhancing the Capabilities of Knowledge Workers

It is clear that large language models will improve the efficiency of knowledge workers. However, it is not yet clear how this will be achieved. Dust is working hard to figure this out. If knowledge managers cannot access proprietary data within the enterprise, they cannot be of much help. Therefore, Dust has built a platform to index, embed, and continuously update the enterprise's internal data (Notion, Slack, Drive, GitHub) and make it available to products supported by large language models.

Dust's co-founders Gabriel Hubert and Stanislas Polu sold a company to Stripe and worked there for 5 years. They witnessed firsthand how rapidly growing companies struggle with scale issues. They saw the so-called "information debt," and now they are focusing on using large language models to address some of the major pain points associated with it. Currently, Dust is exploring the following applications on its platform:

Answer engine: Focused on factual information, as it is crucial for widespread adoption.
Generation assistant: Provides templated assistance in creating content. For example, generating missing paragraphs based on internal data.
Document auto-updating: Document owners receive notifications and suggestions whenever information that should be updated appears within the company.
Structured event extraction: Users can generate structured events from unstructured data (such as Slack threads) based on predefined templates.
Internal data monitoring: Intelligent rules are used to monitor enterprise data. For example, receiving alerts if personally identifiable information (PII) inadvertently appears where it shouldn't.

While the content is extensive, the founders of Dust believe that most of these data flows will eventually converge into a unified product. They are still in the early stages of exploration, shaping the ultimate vision of Dust. Based on initial iterations, they believe they have confirmed their core assumption: the capabilities of knowledge workers can be enhanced (rather than replaced) by large language models, and a new "team operating system" can be built on this basis. — Konstantine Buhler, Partner, Sequoia Capital

6. Labelbox

Unleashing Business Data

"The rise of big data" has been ongoing for over 20 years, and while enterprises continue to ingest more data than ever before, many still struggle to derive insights from this data for AI models. Data processing and annotation remain the most tedious and expensive parts of the AI process, but they are also the most crucial for high-quality results. Even with the increase in pre-trained large language models, enterprises still need to focus on using their proprietary data (across multiple modalities) to create generative AI with unique advantages, providing differentiated services and insights, and improving operational efficiency.

Labelbox simplifies the way enterprises input datasets into AI models, solving this problem. It helps data and machine learning teams find the right data, process and annotate it, push models into production, and continuously measure and improve performance.

The new platform from Labelbox leverages the advantages of generative AI. Model Foundry allows teams to quickly experiment with AI base models from all major closed and open-source providers, enabling them to pre-annotate data and conduct experiments rapidly with just a few clicks. This allows them to understand which model performs best on their data. Model Foundry automatically generates detailed performance metrics for each experiment run and provides version control for the results.

Its impact could be profound. Traditionally, it would take a human several days to complete a simple but time-consuming task, such as categorizing e-commerce listings containing multiple paragraphs. With GPT-4, this task can be completed in a matter of hours. Model Foundry allows enterprises to discover these efficient ways on their own.

This is not the only example. Early results indicate that over 88% of annotation tasks can be accelerated using one or more base models. Labelbox allows anyone to pre-annotate data with just a few clicks, without coding and inputting data into models. This tool is designed to enhance team collaboration and utilize cross-functional expertise to maintain human supervision for data quality assurance. This feature allows language model experts and small to medium-sized enterprises to easily evaluate models, enrich datasets, and collaborate to build intelligent applications, democratizing access to AI.

In fact, Labelbox has significantly reduced costs and improved model quality for some of the world's largest enterprises, including Walmart, Procter & Gamble, Genentech, and Adobe.

For enterprises, it is a competition to unleash the power of these base models on their proprietary data to solve business problems. We look forward to seeing how Labelbox will help enterprises unleash their data to provide better products more efficiently. — Robert Kaplan, Partner, SoftBank

7. Runway

A New Creative Suite

Artificial intelligence is everywhere and is increasingly becoming a commodity. In most cases, companies enrich existing applications with AI as chatbots. Few AI applications reshape the product experience, using this technology to fundamentally change the way we interact with products, much like Google's search engine changed the way we browse the internet, or Instagram changed the way we share photos from our phones. These AI applications require a deep understanding of existing user experiences, visionary product thinking, and cutting-edge technology.

Runway is a leading example of such a company, using applied AI research to reimagine creative experiences and establish a completely new creative suite.

Deep understanding of user experience. Founders Cristobal Valenzuela, Anastasis Germanidis, and Alejandro Matamala-Ortiz are researchers in the Interactive Telecommunications program at New York University with years of design experience. Runway's team understands the creative tool ecosystem from direct experience and popular barriers. For example, creative filmmaking typically requires expensive equipment, software resources, and high-level training. Therefore, creative filmmaking has historically been concentrated in large production studios. Runway saw an opportunity to expand and improve the accessibility of the necessary creative tools.
Visionary product thinking. Runway early on realized that the inflection point of AI could greatly improve user experience, not only enhancing existing creative tools but also fundamentally changing how these tools work. For example, users can create entirely new video content from scratch using simple text prompts. Importantly, this video is professional-grade and shareable on desktop or mobile devices. Regardless of technical skill, background, or resources, Runway can save hours or days of editing labor. This is a visionary product that turns simple prompts into compelling life.
Leading AI technology expertise. Runway not only solves a problem with a visionary product but also reimagines underlying research and technological infrastructure. The company's in-house research is at the forefront of deep neural networks for image and video synthesis. The company has developed Gen-2, a multimodal AI video model more powerful than any product currently on the market. This is the first public model that can convert text into video. Prior to this, Runway released Gen-1, a model that led the paradigm shift in video generation tools, capable of producing high-quality outputs. Runway's researchers also pioneered the Stable Diffusion model for text-to-image.

Since October 2022, Runway has developed over 30 AI "magic tools" covering video, images, 3D, and text, serving various aspects of the creative process from pre-production to post-production. Their client base includes Fortune 500 and Global 2000 companies such as The Late Show with Stephen Colbert, New Balance, Harbor Picture Video, Publicis, and Google. The platform has also been used to edit Oscar-nominated films, such as the Hollywood hit "Everything Everywhere All at Once."

The most exciting AI applications are changing existing product experiences and rethinking how users interact with products. With Runway, users can create new videos in seconds, whether they are first-time video shooters or professional production studios. This is a revolutionary shift and an example of how AI is reshaping different industries. — Grace Isford, Partner, Lux Capital

8. NewLimit

Reshaping Cell Fate

Cells are the most complex computer systems on Earth. Like computer chips, DNA is composed of basic units that produce complex functions. Unlike code based on "bits," atomic-based code has randomness and hierarchy. One system depends on another, and that system depends on other physical systems, each affected by heat, acidity, and molecules in the cellular microenvironment.

Despite these interdependent relationships, cellular machine code (DNA) can effectively run different programs. Although your liver cells and skin cells contain the same genome, the appearance, sensation, and function of these cell types are different. Why? Because they execute different epigenetic programs.

In 2006, Takahashi and others reprogrammed mature cells into stem cells using a combination of four transcription factor (TF) proteins, pioneering the field of epigenetic reprogramming. Transcription factors are proteins that regulate genes, essentially altering the "program" in operation. Takahashi and Yamanaka's discovery led to the birth of induced pluripotent stem cells (iPSCs) and earned them a Nobel Prize. Since then, many research groups have begun applying unique TF combinations to change cell states, rejuvenate damaged cells, and restore youthful cell phenotypes.

While epigenetic reprogramming has become more manageable, it is still no small feat. Research groups must discern which TF combinations can effectively transform cells from state A to the desired state B. For example, future TF combinations may allow us to transform diseased cells into healthy cells, leading to the development of a new class of drugs. We need large-scale reprogramming screening because many application areas do not yet know the exact combinations of TFs. There are over 1,500 native TFs in humans, so a more efficient search method is needed. We believe that NewLimit is designing such a method.

Driven by advances in single-cell sequencing and machine learning technology, NewLimit is transforming previously manual disciplines into data-driven science. The company has a healthy division of labor between molecular biologists and computational biologists, laying the necessary cultural foundation for building an increasingly efficient closed-loop platform. Combining expertise and multimodal readings (scRNA-Seq, scATAC-Seq, etc.), NewLimit aims to discover therapeutic reprogramming factors to treat previously incurable diseases.

In each round of experimentation, NewLimit uses machine language technology to:

Merge and compress multiple detection readings into a low-dimensional optimization space containing the current cell state A and the desired state B.
Enumerate new TF combinations along the optimization space that could push the unit toward its desired state.
Suggest which types of data will help improve the model and when/where to apply more expensive, lower-throughput experimental methods.
Propose changes to the platform to maximize the useful information generated from every dollar.

In addition to its outstanding team, technical prowess, and ambitious vision, we also admire NewLimit's pragmatic approach. While the company has not publicly shared the details of its initial business strategy, we believe this approach is creative, reasonably de-risked, and potentially transformative for humanity. The founding team unanimously believes that platform biotechnology may be likened to expensive scientific projects that do not yield short-term assets. For this reason, NewLimit has remained transparent since its inception and has cataloged its technological progress.

We should be humble in the face of the complexity of nature. It is certain that biology is more difficult to program than the silicon devices we design ourselves. Dimension's goal is to help entrepreneurial pioneers like NewLimit explore the boundaries of possibility at the intersection of technology and biology. — Simon Barnett, Director of Research, Dimension

9. Poolside

Foundational AI for Software Development

OpenAI focuses on general artificial intelligence, DeepMind focuses on scientific discovery, and the third fundamental use case of AI is understanding and creating software.

GPT-4 is deeply entrenched in the workflows of experienced and novice developers. But this pattern shift is still in its early stages. From the past few months, it can be inferred that AI-assisted programming will soon be ubiquitous. As this trend further develops, natural language will become the abstract foundation for building software.

While other companies have also released large pure-code models similar to StarCoder, none have come close to the performance of GPT-4. I believe this is because models trained only on code cannot produce powerful software development capabilities. This is how I came to know Poolside. The company was founded by Jason Warner, former CTO of GitHub, and Eiso Kant, former founder of source{d}, the world's first company to research AI for code.

What sets Poolside apart is that they adopt the OpenAI base model approach but focus only on one function: code generation. Their technological strategy depends on the fact that code can be executed, allowing for immediate and automatic feedback during the learning process. This makes reinforcement learning through code execution possible, a compelling alternative to reinforcement learning based on human feedback (RLHF). This is something Eiso began exploring as early as 2017.

While the potential of beneficial general artificial intelligence (AGI) for humanity is undeniable, its realization is still distant. So why wait for AGI? By focusing on advancing AI in specific domains, such as software development, we can dismantle more creative barriers. I look forward to the day when the Poolside team can realize their vision of building dedicated software base models. — Matan Grinberg, Co-founder and CEO, Factory

10. Mistral

France's Competitor to OpenAI

Recently, the explosive growth of projects in the field of generative artificial intelligence has illuminated Paris. Perhaps you might ask why? My thought is that Paris has the largest pool of world-class talent in the field of generative artificial intelligence, and these talents are still outside the purview of OpenAI. Among these projects, the most daring is undoubtedly Mistral. Mistral, founded by Guillaume Lample, Arthur Mensch, and Timothée Lacroix, has a mission to build the best open-source language models and establish a thriving ecosystem around these models.

I have known Guillaume for four years, and we have both been deeply involved in the application of large language models to the field of mathematics, especially formal mathematics. During our time at OpenAI and Meta, we developed a friendly competitive relationship. Guillaume is one of the most talented researchers I have had the privilege of working with, and I have had the privilege of witnessing his journey from research at Meta to the founding of Mistral. In this process, I also met Arthur Mensch. His work has left a deep impression on me, especially Chinchilla, which redefined the meaning of efficiently training large language models, and RETRO, a method for retrieval-enhanced language modeling that, in my opinion, has not been fully explored.

Now, let's delve into why Mistral has become Mistral. The vision of this startup is to establish an ecosystem based on first-class open-source models. This ecosystem will serve as a launchpad for projects, teams, and companies, accelerating innovation and creative use of large language models.

Take reinforcement learning based on human feedback (RLHF) as an example. Typically, RLHF requires a lot of time and is therefore costly. It involves manual "labeling" of AI actions, which can be a significant amount of work. This effort is only worthwhile when the prospects of the AI model are good enough. For large companies like OpenAI, it makes sense to invest in this process, as the company has the resources to achieve this goal. However, traditional open-source communities often need a "leader" to step up and take on this responsibility.

Mistral has the opportunity to do just that, investing in RLHF on open-source models. By doing so, Mistral will open the door to a Cambrian explosion of innovation. Open-source developers will have the opportunity to access well-labeled models that they can adjust and customize according to different needs. The ultimate winners will be the broader market, and we will have the opportunity to obtain more specific and compelling use cases than those produced by a single closed company.

Whoever has the best open-source models will be better able to attract interest and value. I am optimistic about Mistral because the team is actively pushing the frontier of efficiency/performance. At the same time, Mistral has the most talented people in this regard globally.

Mistral has already secured the team and resources to execute this initial vision. The company has also found partners to evaluate these models in enterprise use cases. Keep a close eye on Mistral, as they are ready to compete with OpenAI. — Stanislas Polu, Co-founder, Dust

11. Sereact

Smarter Industrial Robots

We often hear predictions that, in the long run, artificial intelligence and robotics technology will enhance or automate human tasks. Today, this has become an increasingly urgent business task.

By 2030, the eligible labor force in Europe is expected to decrease by 13.5 million, and labor costs are rising at the fastest rate in over 20 years. With the rise of e-commerce, warehouses are under more pressure than ever, making it increasingly challenging for businesses to remain competitive.

55% of warehouse operating costs come from order picking, but for companies looking to transition to automated systems, the situation is not optimistic. The various flashy applications in AI-led SaaS (Software as a Service) that we are familiar with, or the plethora of open-source products we see in other parts of the ecosystem, have not yet been applied to the robotics field.

Instead, companies seeking automated picking and packing face the choice of expensive, inflexible robot solutions. They must grapple with a large number of proprietary interfaces, which require a significant amount of programming time and expertise. These systems are also difficult to handle changing product assortments, require regular manual intervention, and perform poorly in extreme situations.

Secret has solved these challenges. Its software is based on a powerful simulation environment, training robotic arms to understand the spatial and physical nuances of any potential real-world environment. After deployment, the system will be optimized through continuous learning from real-world data. This also means they can handle the challenge of grasping traditional high-difficulty items such as electronic devices, textiles, fruits, tiles, and wood.

Most excitingly, their robot stack uses large language models to achieve intuitive natural language control of the robots. They have developed a transformer model called "PickGPT," which allows users to give commands and feedback to the robots through voice or text. This means that anyone, regardless of their technical knowledge, can request the robot to perform the desired tasks.

Secret combines the two professional fields of its co-founders. CEO Ralf Gulde has conducted research at the intersection of artificial intelligence and robotics technology, while CTO Marc Tusher specializes in deep learning. Both have conducted peer-reviewed research in these disciplines at the University of Stuttgart, one of Germany's most prestigious universities for automation and industrial manufacturing.

Despite being a young company, Sereact has already attracted numerous impressive partners, including Daimler Truck, Schmalz, Zenfulfillment, Zimmer Group, and Material Bank. This indicates a huge potential market opportunity in the picking and packing industry.

In addition to the obvious applications in e-commerce warehouses (whether it's picking orders or unpacking), there are a range of other use cases. For example, in traditional manufacturing, there is a time-consuming process called assembly, which involves laboriously collecting the fine parts needed for assembly. For a long time, robot arms have struggled to grasp small parts and have difficulty sorting individual parts in a cluttered environment. Sereact's software can identify these parts and select the correct gripper to pick them up.

The Sereact team not only has exceptional technical expertise but also a keen understanding of their customers' work environments and a genuine desire to help customers overcome labor shortages and achieve efficient, sustainable operations. As the first to combine large language models and picking and packing, from academic possibilities to real-world impact, I am confident in their ability to execute and expand as true challengers in the field of robotics. — Nathan Benaich, General Partner, Air Street Capital

12. Lamini

Customized Large Language Model Engine

Now, every business is trying to integrate artificial intelligence into its operations. The world's largest companies have recognized the potential of artificial intelligence, with 20% of CEOs in the S&P 500 mentioning artificial intelligence in their first-quarter earnings conference calls. Large language models can significantly improve business efficiency by accelerating core functions such as customer support, external sales, and coding. Large language models can also improve core product experiences by answering customer questions through AI assistants or creating entirely new generative AI workflows to satisfy customers.

Given that large companies often lag in adopting new technologies, we are surprised by the speed at which businesses are beginning to use artificial intelligence for building. It is not surprising that many companies want to build their own AI models and solutions internally. Every company has a proprietary treasure trove of customer data, often a part of its core business moat. These companies believe that there is a risk in sending their most valuable data to basic model APIs or unreliable startups. Even without considering data privacy issues, public large language models like GPT-4 or Claude are trained on open data, lacking customization for enterprise-specific use cases and customer groups.

Some tech companies, such as Shopify and Canva, have already assembled "AI tiger teams" internally, integrating artificial intelligence into various parts of their business using ready-made open-source models. However, most companies do not have the resources or experienced AI researchers to build and deploy proprietary large language models based on their own data. They realize that this wave of artificial intelligence may be a transformative moment for the future of their business, but so far, they have not been able to leverage or control their own AI development.

This is why we are very excited about what Sharon Zhou and Greg Diamos and their team are doing at Lamini. Lamini is a large language model engine that enables developers to easily and quickly train, fine-tune, deploy, and improve their own models, and provides human feedback. This tool provides a delightful development experience, abstracting the complexity of AI models and, more importantly, allowing businesses to build AI solutions on their own data without the need to hire AI researchers or risk data leaks. Last fall, we first collaborated with Sharon and Greg. Since then, we have had the opportunity to support this technically excellent, customer-centric founding team and help them achieve their ambitious vision of changing the way businesses adopt artificial intelligence.

Specifically, deploying private large language models using Lamini provides a wide range of advantages compared to using public solutions. Internal engineering teams handling the construction process can ensure data privacy and have better flexibility in model selection, as well as the entire computing and data stack. Models made using Lamini also reduce illusions, lower latency, ensure reliable runtime, and lower costs compared to ready-made APIs. These performance enhancements come from the Lamini team's decades of research and industry experience in AI model and GPU optimization, as well as core technical insights built into the product.

Renowned startups and large enterprises have already begun deploying large language models internally and with customers using Lamini, and they are very excited about its setup speed, performance, and reliability. In the future, we believe that every business will use AI in its operations and products, but only a few companies will have dedicated AI teams. This startup, Lamini, is creating a level playing field, helping all companies have the opportunity to leverage this transformative technology. Thanks to their recent collaboration with Databricks, companies can now set up and run their AI solutions on existing Databricks data lakes and compute clusters more easily than ever before. — James Wu, Investor at First Round Capital; Todd Jackson, Partner at First Round Capital

13. Factory

Your Coding "Robot"

Currently, if you want a computer to do something for you, you have to translate your ideas into "computer language," a hypertext code that a compiler can understand. To become an engineer, you have to twist your brain thinking like a machine. However, we are reaching a turning point where artificial intelligence can translate human language into code. The transition from human engineers to digital engineers is likely to be one of the most important technological turning points in our lifetime.

We are still in the early stages of this transition. Artificial intelligence tools like BabyAGI and AutoGPT have sparked public imagination. However, while coding assistants like Github Copilot represent progress, they are still very limited, mainly providing auto-completion for ideas already implemented in the code.

Factory, on the other hand, is different. The company was founded in 2023 by former string theorist Matan Grinberg and machine learning engineer Eno Reyes. When I met Matan, I was immediately drawn to his vision: in the future, engineers can delegate annoying tasks and focus on challenging problems, making the process of building things enjoyable. To achieve this, Matan and Eno created autonomous coding "robots."

The robots are AI engineers capable of handling tasks such as code reviews, debugging, and refactoring. Unlike existing products, Factory's robots do not require your hands-on involvement; they can independently review code, handle errors, and answer questions. You can also use the robots to brainstorm and share the workload, just like a junior developer. The robots have robust safeguards, and their intelligence is user-driven, making it difficult for them to "hallucinate" incorrect answers.

Code generation will be one of the most transformative areas in the AI revolution, and Factory has all the necessary tools for success.

Team. Matan is the CEO of Factory, a string theorist from Princeton University who has imagined singularities in black holes. Eno previously worked as a machine learning engineer at Hugging Face, personally handling tedious engineering processes. This is a unique team.
Practicality. While the robots cannot yet perform tasks as excellently as human engineers, they can stand alone and handle the tasks engineers dislike. Engineers can delegate repetitive work to Factory.
Speed. Factory has achieved remarkable progress in just a few months. While others are still imagining AI engineers, Matan and Eno have already begun development. They are rapidly improving this already outstanding product.

The story of human development is a story of unloading repetitive work, allowing us to turn to more complex tasks. When humans invented agriculture, they actually unleashed our ability to build cities. After the Industrial Revolution, we built rockets that could take humans to the moon. The next generation's task is to free humans from menial tasks online, further driving the technological frontier.

When the only limit is human imagination, what will we build next? — Markie Wagner, Founder and CEO of Delphi Labs

Translator: Jane

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。