Greg Badros

Some folks say that - for them - LLMs can handle large contexts of 100,000s of tokens with no degradation of accuracy in their output. Other folks say they see a rapid deterioration when contexts go beyond sizes that are orders of magnitude smaller. How are some able to achieve that with their model inputs while others apparently can't get them to stay on the rails much further beyond 1000 tokens - and often significantly fewer? For 2 years, I asked myself "What are they doing that I'm not doing? What's the secret?", and I tried to reproduce their results, but in vain. Finally, after a lot of research and experimentation, I realised that I was asking the wrong question.

١٩

‏١١‏ ‏تعليق‏

I spoke with Bay Area founders, engineers and product leaders about the rise of “vibe coding” and what it could mean for tech jobs in San Francisco and beyond. My latest for the San Francisco Business Times. https://lnkd.in/eHqkRs6H

٢٢

Exciting Milestone at Forma Prime! I haven’t shared many updates since joining the Advisory Board at Forma Prime, but today marks a significant milestone worth highlighting. Forma Prime is celebrating its first IPO as Infleqtion, a leader in the quantum technology space, officially begins trading. This follows the successful completion of their merger with the SPAC led by veteran dealmaker Michael Klein. It is an incredible time to be involved with a team supporting such "deep tech" transitions. Infleqtion’s work—ranging from quantum computing to high-precision sensing for NASA and the DoD—is at the forefront of the next technological frontier. I’m looking forward to continuing my work with the Forma Prime board as we identify and scale the next generation of industry-defining companies. Read more about the debut on Bloomberg:

١٩

‏١‏ ‏تعليق واحد‏

The loudest critic of LLMs just raised $1.03 billion to prove them wrong. Yann LeCun, Turing Award winner and former chief AI scientist at Meta, just closed Europe's largest seed round ever for his new startup AMI Labs. Backed by NVIDIA, Jeff Bezos, and Temasek. Valued at $3.5 billion before shipping a single product. His thesis is simple and uncomfortable: large language models are fundamentally limited. They predict text. They don't understand the world. They can't reason about physics, causality, or spatial relationships the way a two-year-old can. So he's building something different. "World models" that learn from observing reality, not just reading the internet. The initial targets are robotics and autonomous transport, domains where understanding the physical world isn't optional. Here's what makes this interesting. NVIDIA is backing a bet that directly challenges the architecture they profit most from. Bezos is funding a company that competes with the AI strategy at Amazon. These aren't charity donations. These are hedge bets by people who have access to every metric in the industry. Four months from founding to a billion-dollar raise. That timeline tells you something about the conviction level behind this. Whether LeCun is right or wrong about LLMs, the capital markets are clearly signaling that the current paradigm isn't the endgame. The smartest money in tech is diversifying away from text prediction. Are we witnessing the beginning of AI's next architectural shift, or is this the most expensive contrarian bet in tech history? #AI #WorldModels #ArtificialIntelligence #Startups #DeepLearning #TechFunding #Innovation

١

Idd. LLMs indeed have fundamental limitations. They are indeed very efficient in imitating natural languages, but the model itself is limited when it comes to higher congnitive processes. https://lnkd.in/dix8a-Mm

١

The most recent Queued Up report from Berkeley Lab is out. Steven Zhang and I are proud to be listed as coauthors on this edition. The Interconnection.fyi team worked closely with Joseph Rand and the rest of the LBNL team to provide the data that powers this year's analysis. Supporting this level of research is core to our mission of bringing transparency to the wholesale energy markets. The 2025 edition (covering data through the end of 2024) highlights some significant shifts in the landscape: Active Capacity: 2024 closed with nearly 2,300 GW of generation and storage seeking interconnection. Changing Mix: Active natural gas capacity increased by 72% year over year, while solar and storage saw slight decreases in total queue volume. The Backlog: 408 GW of capacity already has an executed or draft interconnection agreement but has not yet reached commercial operations. Timelines: For projects built between 2018 and 2024, the median duration from request to operation has doubled compared to the early 2000s. This report is the definitive annual benchmark for the industry and provides a vital baseline as we begin to see the implementation of FERC Order 2023. The analysis in this report is based on our EOY 2024 data snapshot. Since then, we have continued to track and update these queues every single day. If you want a live view of how these numbers have shifted in the months since this snapshot was taken, follow Interconnection.fyi and subscribe to our Substack. You can find the full slide deck, interactive maps, and raw data files at the link in the comments.

٤٢

‏٣‏ ‏تعليق‏