I’m personally looking forward to talking with dolphins

coding with sonnet 4 is nothing short of magical. one-shots pretty much all my coding tasks. it's like o3 without the cost or latency

Anthropic’s models have consistently been better than the sum of its evals. Best in class at out-of-distribution tasks. Unsurprising to see Sonnet 4 set SoTA.

Software development is more than just coding. Introducing Droids -- the world's first software development agents. 🤖 Starting today, Droids are available for general access. Factory integrates with your entire engineering system (GitHub, Slack, Linear, Notion, Sentry) and

Aaron Harris OpenAI Anthropic “One of Silicon Valley’s biggest contradictions is the love of two diametrically opposed things: The use of pattern-recognition to predict the future… and the obsession with a small number of exceptional successes” andrew chen andrewchen.com/how-sheep-like…

James Campbell Situational awareness was net-negative IMO because it freaked out the Trump admin and led to conversations around banning open-source AI access in the US, which would obviously decimate the industry

gen z founders on why you should join their team:

Today, I’m thrilled to finally share what we’ve been building at Traversal and announce our launch out of stealth—backed by $48M in seed & Series A funding led by Sequoia Capital and Kleiner Perkins, with NFDG, Hanabi Capital, and an incredible group of angels. Traversal is

Reinforcement Learning (RL) is unleashing the next wave of AI capabilities. And it's no longer just a secret tucked away in research labs...🧵 Here are just a few of the startups tackling current RL challenges like task generalization, designing effective evals, and crafting

The Arc deadline is Sunday. This could be your path to partnering with @Sequoia & to spending time with the most ambitious people I know (and me!). DM or apply :)

Arc gave us the opportunity to work with incredible mentors like Thilo, it prepped us with some of the most impactful frameworks that we still apply to this day, and it surrounded us with the most ambitious / impressive builders who I’m still friends with to this day Go

Andrej Karpathy Vincent Weisser This is exactly what we are building at fleet

You know who’s killing it right now? That cohort of founders who dropped out of school in 2020 to start a company that failed / sold for peanuts, and are now onto their second. Only loosely keeping track but a number of them have already blazed past series B and still crushing.

early stage investing is knowing this founder will be successful but hoping it will be with this company

all model companies were pretraining on the ~same internet. of course, grok has access to twitter dataset and gemini can pretrain on youtube & so on. but it's mostly the same internet otoh, rl envs will be w.e the lab chooses to prioritize. so you should expect more speciation

Teams can tell the difference between leaders driven by paranoia and those driven by fear. Paranoia is rooted in the legitimate concern that everyone is out to get you in business, and aids practically and directly in strategy. Fear is reactionary, panicky, and unpredictable.

The best founders I know are all questionable at best at the 0-1, but absolute killers at the 1-n. If you’re charismatic and can push things through no matter the cost, you end up with false positives that may continue to lead you down faulty paths.

What does Applied Compute actually do? What does the data that Mercor actually sells look like? In software land, evals and tool use capabilities are the last frontier to conquer before we can design models for enterprise use cases. To fuel this, we need engineering real world