Moe Khalil

@WhiteHouse Lmao what

They walk among us. ALIENS.GOV

Lmao what

@toddsaunders Tbh haven’t personally felt a difference since 4.5, but maybe I’m just using it wrong

If you are using Opus 4.8 today, what’s the biggest noticeable difference to you vs 4.7? Benchmarks are cool but I want some real examples.

@KOT4Q @notbpod People overreacting right now but this team got put together in February. No reason not to give it one more try.

What’s next for The Cavs?

@catboosted Don’t personally believe in kingmaking but I think Pinecone is a bad example; the world just moved on from RAG + the model providers almost always had an advantage in providing embedding

Does Pinecone prove that kingmaking is dead?

@mstockton Or just use wisprflow

Stepping away from the dopamine chat box and recording a long, rambling, yet intentional voice memo covering everything you want to accomplish in whatever project you’re working on. Then transcribing that and using it as input to the model, where you then ask the model to ask you questions exhaustively until it knows exactly what to do in order to accomplish your goal

Is anyone working on a social media with no bots/LLMs? Seems like an impossible problem but would legitimately 10x any other existing platform it’s getting exhausting.

Tokenmaxxing is antithetical to productivity. You're giving yourself a false sense of security because your "agents are running." Most pre-PMF startups would do more to move their company if they stopped using coding agents for a week and focused on the real bottleneck to growth. It's almost never more code.

@geoffreywoo Also in big part because people enjoy building for themselves. I'm sure if AI was built by a group of historians it's capabilities would look very different.

ai is best at solving coding and math. so does that mean coding and math are less important to study? the answer is a definitive no, because the people who drive ai the best are coders and mathematicians. coders and mathematicians are more powerful than ever.

@keaganwernicke @SuperDuperSF

Who is building super duper intelligence?

Hasn't almost every human discovery been a result of logical exploration or brute force? Not saying I think they will replace humans because you still need someone to ask the questions (s/o Douglas Adams), but as long as we build the right environments for AI to run brute force experiments (coding right now, real world environments in the future) I don't see why they can't continue to make discoveries.

I want to emphasize something Mo said which I think is helpful. LLMs are like a *new species* that’s been discovered; it’s not going to suddenly turn into something else—which is the only way the current chatbot mania makes sense. LLMs are a great tool for exploring massive amounts of data by compressing their relationships into weights through training, and allowing queries through natural language. But they can only ever take from what they’re given. There may be nuggets within the existing human information space, like the solution to the Erdos problem, that it will find through brute force or logical exploration. That’s worth continuing to try. But that’s more like mining for Bitcoin than overseeing something intelligent. These things can be very useful, but they’re just a software feature. They will never be replace people.

There’s already countless examples of such tech in existence. If nuclear weaponry didn’t cause human extinction, I’m not sure how agents living on your computer would.

@ThePrimeagen If you’re pre-PMF, all you’re doing by running more agents is adding more code that will probably end up being scrapped while comforting yourself with the illusion of productivity. If you think you know what you’ll be working on in a year, you’re probably wrong.

Honestly why stop at 100x engineer? Just use more agents, you literally could be 1000x, 10000x, 100000x just by scaling You could what you use to in an entire year in one second

If you’re investing your money without spending at least $50 on deep research first you’re ngmi

@signulll It’s only hell if you don’t actually believe in it. It’s pretty easy to hold unconventional beliefs if you’ve seen something yourself to prove otherwise.

believing in something early is effectively learning to tolerate being alone in a position long enough for the world to catch up, if it ever does. if you do this over & over again, you can imagine why sustaining this is difficult as hell.

Would people be interested in a deep-research focused event in San Francisco? It's an extremely powerful tool that is being way underused and would love to share thoughts with people on what they're using it for.

What's the best consumer agent product out there right now? I've only been using Cursor but looking to try something for non-coding work.

@adamdotdev In our product we made it so that the user sets the budget and the model works for that long. So you control it via spending. It’s felt way more intuitive that way.

Why do I have to tell the model how hard to think, no way that is a thing we have to do long term

I think people are using deep research for the wrong things. There is an audience out there that can write thoughtful prompts (given context), read a lot of text in an instant, and become a more powerful version of itself as a result.

I'm a huge Survivor fan, and I recently got the #Survivor50 winner spoiled by Kalshi. This got me thinking: how many pre-recorded reality TV shows exist where you can make money by intuiting insider information? I ran a $50 Webhound search, and here's what I uncovered. webhound.ai/p/suspicious-u…