Caleb Harris

founder @tryandai, former @mit @ycombinator

[Caleb Harris @caleb_and_ai]

@wilson_spearman @benswerd Correct answer

[Caleb Harris @caleb_and_ai]

really interesting to see more concepts from neuroscience surface as emergent properties in LLMs

[Anthropic @AnthropicAI]

New Anthropic research: A global workspace in language models.

Of everything happening in your brain right now, only a tiny fraction is consciously accessible—thoughts you can describe, hold in mind, and reason with.

We found a strikingly similar divide inside Claude.

[Caleb Harris @caleb_and_ai]

@raunakdoesdev @jxnlco i find i get better results much of the time just using remote codex from my phone

[Will Jack @willwjack]

Whether or not AI recommends you is becoming make or break for many companies.

Sadly, the AEO/GEO space is full of people selling misleading metrics for a quick buck.

I’m really irritated by dishonesty, so we analyzed 37,000 AI conversations to figure out what actually works🧵

[Nyle Sykes @nylesykes]

I met @jasonyuan in 2024 while he was building Dot at @newcomputer. I was struck by his thoughtfulness, authenticity, and deep passion for what he was building, and as I looked at what he’s created over time, it became clear that he’s a special person whose intuition allows him to see the future before it’s obvious.

In 2019, Jason created Mercury OS. He told us the story of how it came together: at the time, he felt, “This computer doesn’t adapt to my brain. I have all these files. There’s so much value from emails. It has Siri. Why am I opening all these apps to do things, to respond to emails, to coordinate?” This led him to envision a computer that understood context and could generate interfaces on the fly – which seven years later, we now refer to as generative UI.

In 2023, he teamed up with @sjwhitmore to build Dot, an AI with deep, personal memory designed to help you better understand yourself. It was an early bet on memory and proactivity before either became central to consumer AI, and I felt it was a beautiful product, both visually and how it felt to interact with.

@hivemind is the culmination of a lifetime of Jason’s insights and intuition and years of experimentation at the intersection of AI and social. And I believe that, once again, he’s seeing something the rest of us haven’t – a new kind of social network (check out the video!). It’s a privilege to have had the opportunity to back Jason when he had nothing but an idea, and to have since seen firsthand so many magical moments in the Hive.

The team is hiring – please reach out to me or Jason if you want to learn more!

[Caleb Harris @caleb_and_ai]

@thesubhashk This is cool

[Caleb Harris @caleb_and_ai]

@jgreze @TownAI @arampell @a16z @kirstengreen @ForerunnerVC @firstround @altcap @conviction Congrats @jgreze!

[Caleb Harris @caleb_and_ai]

@seema_amble so good!

[Caleb Harris @caleb_and_ai]

@bosmeny @ycombinator @sama @dflieb such a good idea

[Caleb Harris @caleb_and_ai]

@esthercrawford 🚀

[Anthropic @AnthropicAI]

New Anthropic research: Natural Language Autoencoders.

Models like Claude talk in words but think in numbers. The numbers—called activations—encode Claude’s thoughts, but not in a language we can read.

Here, we train Claude to translate its activations into human-readable text.

[Caleb Harris @caleb_and_ai]

@AnthropicAI Whoa @thesubhashk this is awesome

[Adit @aditabrm]

Today, we’re setting the new standard for complex document extraction.

Introducing Deep Extract. It utilizes an agent harness approach that repeatedly iterates and verifies outputs until they are at human-level accuracy.

More below:

[Adam Cohen @adambcohen93]

Permanent underclass in engineering?

Elite engineers are now 5x more productive than average ones.

[Raunak @raunakdoesdev]

https://t.co/gQKDSr30sH

[Caleb Harris @caleb_and_ai]

@aditabrm Great advice

[Caleb Harris @caleb_and_ai]

@wilson_spearman @Lovable @antonosika 💀

[Caleb Harris @caleb_and_ai]

@adambcohen93 It already feels like there's been a substantial uptick from January to February with Codex 5.3

[Caleb Harris @caleb_and_ai]

@thesubhashk Latently reasoning about being in an eval is a really interesting finding

[&AI @tryandai]

https://t.co/7FbKtB0DwD