Elliot Bensabat

You realize how big the problem/opportunity in evals is when you see a company like Anthropic releasing Opus 5 thinking it's a good model. It's truly unusable and forces you to just spend your Fable tokens. Also makes me bullish on Agentic Development Platforms (where switching models while coding is easier) and better model routers (where hopefully it could just know to use Opus 5 only for the little things it's good at)

Still surprised this isn’t pushed more. We need a protocol (think open banking) to make it easy and safe for consumers to move/share their data from your llm chats. Would be good for competition and thus for users. @elonmusk and the Grok team should do that. They had to reinvent distribution of that product (the @ grok) to start growth, but if it was easier to switch, i’m sure market share would skyrocket

Is anyone pushing for memory portability across LLMs? More time passes, more lock-in, harder for competitors to compete. OpenAI led early with the chats where memory compounds - workouts, travel, day-to-day life. Anthropic focused more on code and work, which are self-contained. So if Anthropic wants to win on the personal side, pushing for memory portability seems like an important move? cc @DarioAmodei @mikeyk Claude's been better for most of my new projects, but ChatGPT has so much context on my life that switching feels costly.

That's actually the best way to solve the problem of people posting fake ARR numbers/charts. Build a feature on Stripe's dashboard to request a tweet from a Stripe bot (e.g. @StripeARR) for any milestone. The tweet will @ the company/founder with the number and date of achievement. The founder can repost that tweet or refer to it as proof, bringing attention & free marketing back to Stripe. cc @patrickc

This summarizes best why Israel lost the info war in the last 20 years. This combined with the fact that even if debunked, there’s no certainty that 1) the mass will trust the proof - hamas terrorists literally had GoPros recording everything from their chest and some people still don’t believe it 2) the mass will see the proof - people consume more content than ever, and you seeing a propaganda post doesn’t guarantee that you will see its debunking. Though this one feels like a problem that actually could be solved

LinkedIn really is proof of network effects 101. Average platform at best, but still the market leader. Probably one of the worst value/money consumer membership out there for their premium product (you really mostly pay to see who views your profile), while still being massively adopted. Wondering what it will take for someone to compete with it - but guessing someone tackling a niche (e.g. possibly something invite-only for top talent? I could see a need in AI recruiting) really well before expanding

Copilot and Ghostwriter are 2 examples of successful LLM integration in your IDE for better productivity. But I haven’t seen anything yet that takes advantage of notebooks. The step-by-step/cells format seems like the perfect environment for LLMs to recommend the right thing at the right time. Some ideas: -Auto-create cells with visualizations or tests based on the last executed cell -Prompt users for the right next steps (« you just trained your model, nice. Now let’s test it with the below code ») Etc. @jeremyphoward you’re the notebook master, have you heard about any breakthroughs there?

After listening to @saranormous and @eladgil ’s podcast with @c_valenzuelab, I started to think about what could make AI art unique (and thus valuable?), the same way that a painting is. I see 3 things: 1.uniqueness of the input (prompt, model parameters) 2.uniqueness of the model 3.uniqueness of how the AI output was combined with traditional art (e.g. painting) Which brings the question of will the next big AI artist be from a technical background (1 and 2), an art background (3) or both? Will AI artist be able to gatekeep their technical edge or will it be open source? A lot of interesting questions… Curious if you’re following a specific artist in the space?

The weird thing about reading management/business books is that it can constrain your thoughts. Of course you learn a lot by reading them, but I feel like it can block you in the local minima of the writer (whether it’s a good or bad one)

So you can now generate a business idea with chatGPT from @OpenAI and then get the deck ready from @magicaltome in < 2 minutes. The illustrations could be improved, but wow. VCs, get ready to have your inboxes flooded.