Jerene Yang

Large Language Models (LLMs) as super lossy compression. I have been thinking a lot about demystifying what LLMs do because lots of y'all's are very mystified. In the linked article, I try to reframe them for you, hoping it will help you understand what they good for and what they are not good for. Side note: I am still looking for my next adventure. If communicating complicated concepts so that regular people can grasp them is something you need, you should message me. I write code and stuff too and am usually fun to be around. #WrittenByMe

6

3 Comments

Waymo gets $5B additional investment. That's an impressive commitment in the current investment client (albeit Alphabet money, not external money). I hope that this gives them enough runway to continue their previous historical moderate pace of expanding instead of trying to go exponential any time soon. We've been seeing some more concerning Waymo mishaps lately that might (or might not) be attributable to pressure to show faster results. Premature rapid scaling didn't work out well for Cruise. Waymo needs to avoid that pitfall. https://lnkd.in/eW4UJVSx

110

11 Comments

Waymo is the best tech product I tried in years, and I can't wait for humans to be banned from driving. What's Waymo? Waymo has been at the forefront of autonomous vehicle development since its inception as the Google Self-Driving Car Project in 2009. The core of Waymo's technology lies in its sophisticated combination of sensors, including LiDAR, cameras, and radar, integrated with artificial intelligence to perceive the environment accurately. This technology allows the vehicle to make quick decisions in real-time, such as identifying pedestrians, bicyclists, and other vehicles, understanding traffic signals, and adapting to unexpected situations like road work or detours. Over the years, Waymo has expanded its testing grounds from a single city in California to multiple locations across the United States, including San Francisco, Phoenix, and Detroit. In addition to testing, Waymo has launched pilot programs like Waymo One (The service I tried), a self-driving taxi service in San Francisco, Phoenix, and Los Angeles. You can use the app to request a ride in a self-driving vehicle. This service marks a significant milestone in commercializing autonomous technology, offering a glimpse into the potential future of urban transport. Additionally, Waymo Via focuses on goods delivery, utilizing the same technology to transport products efficiently and safely, which could revolutionize supply chain logistics by reducing costs and improving delivery times. I am grateful to Amjad for the access to Waymo One. Which city in the Middle East will be the first to welcome Waymo? I would say Dubai, what about you? #selfdriving #waymo

22

9 Comments

Unique role in a unique company! *Board Chair position* for one of the leading teams in generative machine learning committed to designing proteins - https://www.dreamfold.ai/ If that's you or someone you know - please reach out! About the Company: * DreamFold’s mission is to find treatments for severe, life-threatening diseases.  * DreamFold originated from the labs of renowned AI experts Yoshua Bengio (Turing Award winner) and Michael Bronstein (co-founder & DeepMind professor at Oxford).  * DreamFold has established a strong publication record: FoldFlow and  FoldFlow-2 are state-of-the-art models for protein structure design.  * With initial funding from top-tier deep tech investors, DreamFold is now preparing for a funding round it aims to close in Q3 2025. Objective: Work closely with the founding team and investors to help position DreamFold as a leading company in generative protein design. Bring expertise, insight, and capabilities to build a biotech company with an innovative modality and platform. Year 1: Prepare DreamFold for fundraising by supporting the CEO and team in fine-tuning the value proposition.  Year 2: Complete the Series A funding round to help scale DreamFold and establish two or more significant partnerships. In addition to the usual responsibilities of a Chair of the Board, the candidate will:  * Work with the founding team and investors to establish and articulate a clear, investable vision, strategic roadmap, and plan for sustainable growth.  * Take a leading role in promoting DreamFold’s assets, people, and technologies to the stakeholder community (e.g., biopharma, investors).  * Help develop the skills and competency base by mentoring and recruiting the initial advisory board. Attributes: The founding DreamFold team and investors are ambitious and strongly believe they can contribute to developing new treatments. Chair candidates should possess the following attributes:  * Ideally, the candidate has been a founder of a biotech company that scaled through multiple rounds of funding and deployed innovative technologies in biologics or novel advanced modalities to the clinic.  * Has personally led a company through value-enhancing milestones and returned value to shareholders through a meaningful liquidity event.  * Enthusiastic about the potential of generative algorithms in protein/drug design and willing to collaborate closely with the technical team.  * Able to commit to one day a week initially. #ai #genai #board #recruiting #biotech #pharma

13

1 Comment

Large language models are wonderful at ingesting large amounts of content & summarizing. Uploading an academic paper and I can pester it with an infinite list of questions & it will respond with equally infinite patience. In comparing the two most recent Microsoft earnings calls, Claude highlighted: - faster than expected Azure growth (29% vs 27%) - AI contributing 8% of Azure revenue up from 7% - higher CapEx spending & greater capacity constraints for data centers - much better commercial bookings growth : 17% vs 13% Excellent analysis. Benn Stancil described LLMs as great averagers of information. But what if I don’t want the average? If I seek the data point two standard deviations out? If I’m channeling my inner Anthony Bourdain & I’m seeking the fermented shark or cobra heart within my query? At Google, we ranked web pages & ads with many signals. An engineer taught me that of them found rarity of a word across a set of documents. For example, if there’s a document containing 10 instances of the word ghotiacross a collection of documents, where the average document has 0 ghotis, that document is likely the best to answer a search about fish. I haven’t found a way to goad an LLM to produce the rare result. Why did one user decide to use a particular piece of software when 10 others did not? Maybe I haven’t learned how to prompt an LLM well. Ideogram launched a feature called Magic Prompt that expands a basic prompt into richer instructions that elicit better results, marrying the language of a user & a computer. A user might write : A boy with a dog in a park. Magic Prompt replies : A heartwarming scene of a young boy playing with his happy, wagging dog in a lush, green park. The boy has a playful smile on his face while he tosses a bright yellow ball for his furry friend to fetch. The park is filled with beautiful flowers and trees, creating a serene atmosphere for this bonding moment between boy and dog. Magic Prompt surprised me in three ways. First, unlike Google search where terse queries tend to work very well, LLMs relish verbosity. Less Hemingway & more Faulkner. Second, maybe there’s a prompt for rarity. If you know what it is, please share with the rarity shibboleth. Third, I wonder what other prompt techniques are out there that have yet to be discovered to wrest the best from an LLM. This technique is called TF-IDF. It means Term Frequency Inverse Document Frequency.

21

6 Comments

Some takeaways and opportunities for new AI product tooling from my time at KubeCon last week: 🖥️ CPU Inference and On-Prem Deployments: While we're still in the early innings of GPU Inference optimization, there are opportunities worth building towards in the CPU inference space as well. Optimizing CPU inference on-premises is critical for reducing cloud reliance, improving data privacy, and cutting costs. Companies aim to streamline local compute efficiency by enabling faster, more resilient operations with reduced network dependency. 🎮 Texture Generation and Modeling Behavior in AI Systems: Advanced AI techniques, such as texture and mixed-generative models, are proving useful in applications requiring nuanced behavioral analysis and are particularly valuable in industries like gaming and simulation, where creating realistic environments and interactions is crucial. However, training these AI models often requires vast datasets and considerable processing power. Infrastructure teams must ensure that these models can operate at scale, often requiring hybrid solutions that span both on-prem and cloud environments. 👀 Observability and Real-Time Monitoring on multi-format data: AI-powered observability tools can aggregate and analyze systems across data from kernel, applications, and infrastructure layers and do so in real-time, improving troubleshooting and lowering MTTR (mean time to response). However, balancing insights with rising data storage costs remains a key challenge for enterprises. 💻 Modernizing Codebases with AI Transformation Tools: AI tools are helping modernize legacy codebases, reducing technical debt and easing maintenance. While adoption challenges persist, these tools are evolving to make code updates faster and more efficient. Solutions like Moderne, and Codemod are enabling engineers to translate outdated code into modern, efficient formats with minimal manual intervention. 💡 AI-Driven Cost Management and Efficiency: Infrastructure costs are continuing to skyrocket as companies scale up, particularly in terms of data storage, compute, and network usage. AI offers promising opportunities to optimize these expenses through predictive cost management, dynamic resource allocation, and automated scaling. By implementing tools that forecast usage trends and recommend cost-saving measures, such as with CloudZero organizations can gain a clearer picture of their resource consumption. This proactive stance on cost management allows for more efficient resource utilization, ensuring that infrastructure spending aligns with actual demand. AI is no longer a nice-to-have in the infrastructure software space; it’s quickly becoming a competitive necessity. If you're a founder building in any of these areas or themes, I'd love to chat.

41

3 Comments

At ScreensAI, we're focused on driving as much accuracy as we can with LLMs. But for contract review and playbook execution, it's often the case that vague instructions and subjectivity are as much to blame for undesirable results as the LLM's reasoning ability. Say you want to use an AI system to help determine if the following standard is met across thousands of contracts in your repository: 📜 The contract should limit the vendor's liability to 12 months’ fees. From our recently published Screens Accuracy Evaluation Report: "While any given individual contract reviewer may have a reasonable understanding of what they’d expect to pass or fail this standard, we see many edge cases in practice. If preferences for those edge cases are not clearly articulated, the LLM is left to assume and make a judgment call. Consider these scenarios: - What if there are major exceptions to this liability cap that weaken it so substantially that it might as well not exist? - What if this liability cap doesn’t apply to beta features? Or some specific set of features? What if the service is free? - What if the liability is capped at either 12 months’ fees or some fixed amount, whichever is greater? Or whichever is lesser? What if the fixed amount is very large? Or trivially small? - What if there is a secondary cap or a backup cap that applies under certain conditions at a much different amount than 12 months’ fees? - What if the cap is only for fees paid for the services that gave rise to the liability rather than fees paid in general under the agreement? In any of these cases, the LLM is left to assume. The leading LLMs are capable of reasoning through this type of nuance, but if they are told they must make a decision as to whether or not the standard passes, they might not make the same decision that you would if you don’t expose your preferences." 📊 If you want quantitative aggregate results to assess the risk of a portfolio of contracts, thousands of verbose "it depends" responses won't cut it. Contract professionals have strong opinions, institutional knowledge, and well-defined playbooks. ScreensAI helps them articulate this to the platform up front, systematically validating the types of LLM-powered results they'll get for the given set of expressions of their preferences. Then they can deploy this expression of preferences at scale in perpetuity. 🧙‍♂️ We think this is where a lot of the magic happens in LLM-powered contract review.

21

5 Comments

AI memory is fundamental for conversational AI and AI agents. It is critical for building personalized experiences and improving reasoning. Check out Letta’s course to understand how to use AI memory to enhance the customization and quality of your AI app.

15

"AI is powerful, but without the right data foundation, it’s just noise." In the latest Tech Teams Today podcast episode, I had a great conversation with Yongsheng Wu, VP of Engineering at Granica, about why a solid data foundation is crucial for AI success. Here are the top insights from the episode: 1️⃣ Data Foundations Matter: Before diving into AI, ensure your data systems are robust. Solid pipelines & data integrity are critical for meaningful insights. 2️⃣ AI as an Enhancer: AI and ML should optimize existing processes, not solve undefined problems. 3️⃣ Data Quality is King: Curating and labeling data accurately is vital for effective AI models. The better the data, the better the outcomes. 4️⃣ Efficiency & Cost-Effectiveness: Successful AI solutions balance effectiveness with cost efficiency, especially as models scale. Catch the full episode on all major podcast platforms or watch it on YouTube: https://lnkd.in/dbkE__JC #AI #EngineeringLeadership #DataScience #MachineLearning #TechTeamsToday

30

4 Comments

Prompting for scale will soon be obsolete... i think...i hope. Prompting is an error-prone, discovery-based process, unsuitable for scalable AI. Many folks rely on black-box LLMs behind API walls (e.g., Claude, GPT-4), lacking direct control over LLM model weights. Programmatically custom-pruned LLMs will be the future of controllable, production AI, not prompting. hint--> #steeringvectors

17

6 Comments

Very solid lecture on the absolutely mind-blowing progression of LLMs, even though the transformer came out in 2017 the last couple years have been crazy for open source model research. Learn about how models are made useable. https://lnkd.in/gsfX7BCG

8

Everyone knows Moore’s law, the idea that CPU performance doubles roughly every two years. But remarkably, driven by GPU performance and AI model breakthroughs, we’ve seen a very different law emerge with respect to the rate at which AI model performance improves — in many cases improving by orders of magnitude in the past decade. This introduces massive implications in a world of AI Agents. For the past five plus decades, in any domain that computers touch, we’ve generally seen massive efficiency gains that bring more tech to more people. Yet the same has not been possible for essentially any non-digital services. Historically, anytime you want to solve non-computerized business problems (i.e. most knowledge work), we have seen the cost to solve a problem — say from healthcare services to legal work — tend to only go up over time. Due to a mix of inflation, specialization in certain fields, regulation, decreased competition in markets, or other added costs, we’ve seen the price of most services generally monotonically increase over time. This also puts a burden on driving economic activity in general, especially when creating new ideas from scratch, or starting and running a new business. Yet, in a world of AI Agents, anyone — from a small business or large enterprise — has access to automated work that can let them address problems with variable capacity. When you decrease the cost of entry for getting knowledge work done, you dramatically increase the use cases for that work. That new sales program can be tested instantly instead of waiting weeks or months or never happening at all; you can test a broader surface area of your product for bugs or security issues; legal work and reviews becomes affordable on the seemingly insignificant transactions that otherwise produce risk; or you can launch a new marketing campaign in markets you didn’t otherwise serve. Even in a world of AI costs remaining stable this is a super compelling proposition, but in a world of constantly improving technology the implications are vast. Just as Moore’s law steadily lowered the cost of software, increased its sophistication, and made it more ubiquitous, we can imagine a world where knowledge work regularly gets more affordable, sophisticated, and ubiquitous at a constant rate. This means AI Agents present the ability for any unit of work to continuously drop overtime, enabling us to solve greater and greater problems for a wider set of people and businesses. Of course, a large portion of these efficiency gains will quickly get “eaten up” by solving increasingly more sophisticated problems, but those problems will be solved at the cost of a previously simpler problem. The implications of this are enormously interesting, and it’s going to create incredible opportunities all around.

150

13 Comments

Worth reading: Generative AI is not going to build your engineering team for you. So, so good - peppered with gem quotes and observations- chuckle. https://lnkd.in/gZBKtU4X Thanks to Eric Brooke for sharing.

16

**Day 2 as a Headstarter AI fellow with Jack O'Brien** Today, I had the pleasure of being introduced to Jack O'Brien (former Google software Engineer and tech lead, & former CTO of A16z-backed company “stonks”). In todays meeting Jack O'Brien shared valuable insights on how to acquire your first users, build a waitlist, and generate your first $1,000 in revenue. Here are some key takeaways. 1. **Create something cool enough**: Develope something as little as a landing page to get people to sing up and test if people would actually use your product. 2. **Leverage your network**: Get friends, family, and acquaintances to use your product and provide feedback. 3. **Promote Widely**: Share your product all over social media to gain visibility and attract users. A huge thank you to Jack O'Brien for the insightful session and to the Headstarter AI team Faizan Ahmed Yasin Ehsan 🚀 for organizing another great day of learning!

50

1 Comment

✦ ANYSPHERE, the developer of CURSOR, experiences a SURGE in REVENUE growth, drawing interest from PROMINENT venture capital firms. ✦ UNSOLICITED BIDS from notable firms like BENCHMARK and INDEX bring Anysphere's VALUATION to $2.5 BILLION, an increase from $400 MILLION. ✦ Founded by MIT STUDENTS, Anysphere leads in AI-POWERED coding assistance, positioning itself as a POTENTIAL THREAT to traditional developer roles. #Anysphere #TechInnovation #AIAdvancements

2

1 Comment

Very useful insights for folks who are starting to dig further and further into the realm of LLM. An easier GUI approach for fine tuning LLMs.

9

AWS CEO: "Graviton 4 absolutely outperforms the best other processors, x86 processors, at a 20% lower price. So many of our customers get 40 to 50% price performance gains while also using less power and improving their carbon footprint using Graviton. It's because we control that whole process" #aws #artificialintelligence #llms #chatgpt #generativeai #gpu #servers #datacenter #cloud #ai #technology #innovation #business #Semiconductorindustry #electronics #semiconductor #supplychain credit: the transcript

13

3 Comments

After Apple’s research saying LLMs can’t think, it’s funny that now all I’m seeing is research on teaching them to think – something most humans still struggle with. Papers: Looking Inward: Language Models Can Learn About Themselves by Introspection https://lnkd.in/gf6rgnUP On The Planning Abilities of OpenAI's o1 Models: Feasibility, Optimality, and Generalisability https://lnkd.in/gNsPtDbi Thinking LLMs: General Instruction Following with Thought Generation https://lnkd.in/guQtRnsD #AIResearch #MachineLearning #ArticificalIntelligence

24