Brian Shultz

"You're absolutely right! That was a cliff, not a shortcut. Calling emergency services now! Is there anything else I can help with?"

@karrisaarinen @thdxr 2021 but with agents...I wonder what the founder mode phase will look like

@thdxr The new headcount metric

there's been this funny trend where companies flex how much they spend on AI to signal how lucrative their work is "any amount is worth it" it's gonna flip to flexing how conservative they are with AI to signal how costly issues/downtime is for them

@jasonlaster11 This would be incredible. Would cut through so much OKR theater 😆

Should teams deploy internal prediction markets tied to top line company KPIs / OKRs / Roadmap trackers? Increased accountability feels like a valuable muscle.

@jasonlaster11 This would be incredible. Would cut through two much OKR theater 😆

@ConorMcCarter “The first penguin”

OpenAI partnering w Cerebras is the most underrated thing to happen this year. Cerebras is the fastest inference provider in history. Like 50-100x faster. Every response is fucking *instant*. Now imagine running Codex CLI with a SOTA coding model at 5K tokens/sec... openai.com/index/cerebras…

@brian_lovin 🤣

Get yerself some friends who give you Real feedback.

Appreciate the grounded POV! Seems like the main reason Claude Cowork isn’t as useful is that knowledge work isn’t local like your codebase is. A lot of jobs live in the browser/cloud and requires MCP/extra gymnastics to get the right context. Just isn’t as elegant of an experience unless your work context is local

Every re-post on this: "OMFG going to change the world, share it with everyone including your mom and your dog walker, it's so over." Me: "Here is a step by step walkthrough of trying to use it for basic use cases and my honest thoughts." Like & subscribe for honest takes 🫶

Aerospace engineers are the next AI researchers, except even more scarce

100% certainty that @elonmusk rounded up from $420.69

The best experience deletes friction.

Calling it now: Google Calendar will let you auto-book a @Waymo for every event you travel to.

@shensi @merge_api 🔥

haters said that @merge_api couldn't expand beyond hr + recruiting. this year, we closed the largest ai company in the world, the largest bank in the world, the largest payments company in the world. and just now, we closed the largest streaming service in the world. for an internal use case. context is king, and merge is becoming the industry-standard context and integration layer. so proud of this team.

this is fucking nuts Someone emailed 165 support emails at once (including Tango’s) and now they are all auto replying to each other over and over again infinitely It’s only been like 10 min and I’ve received HUNDREDS of emails 🤣

@karpathy And the most verifiable tasks are those that can be verified deterministically

Sharing an interesting recent conversation on AI's impact on the economy. AI has been compared to various historical precedents: electricity, industrial revolution, etc., I think the strongest analogy is that of AI as a new computing paradigm (Software 2.0) because both are fundamentally about the automation of digital information processing. If you were to forecast the impact of computing on the job market in ~1980s, the most predictive feature of a task/job you'd look at is to what extent the algorithm of it is fixed, i.e. are you just mechanically transforming information according to rote, easy to specify rules (e.g. typing, bookkeeping, human calculators, etc.)? Back then, this was the class of programs that the computing capability of that era allowed us to write (by hand, manually). With AI now, we are able to write new programs that we could never hope to write by hand before. We do it by specifying objectives (e.g. classification accuracy, reward functions), and we search the program space via gradient descent to find neural networks that work well against that objective. This is my Software 2.0 blog post from a while ago. In this new programming paradigm then, the new most predictive feature to look at is verifiability. If a task/job is verifiable, then it is optimizable directly or via reinforcement learning, and a neural net can be trained to work extremely well. It's about to what extent an AI can "practice" something. The environment has to be resettable (you can start a new attempt), efficient (a lot attempts can be made), and rewardable (there is some automated process to reward any specific attempt that was made). The more a task/job is verifiable, the more amenable it is to automation in the new programming paradigm. If it is not verifiable, it has to fall out from neural net magic of generalization fingers crossed, or via weaker means like imitation. This is what's driving the "jagged" frontier of progress in LLMs. Tasks that are verifiable progress rapidly, including possibly beyond the ability of top experts (e.g. math, code, amount of time spent watching videos, anything that looks like puzzles with correct answers), while many others lag by comparison (creative, strategic, tasks that combine real-world knowledge, state, context and common sense). Software 1.0 easily automates what you can specify. Software 2.0 easily automates what you can verify.

@MaximeHeckel Such insane discipline 🙇🏻‍♂️

750/750mi ran this year 110 runs 44 weeks 0 weeks skipped 0 min of music listened during said runs 0 injuries 0 dog bites (iykyk) very happy and proud about the consistency and how i managed to push through even when it was too hot/cold/rainy onto the next 1000mi!

LA is so hooked on Waymo that on the first day of rain, we broke the app

@MaximeHeckel this is so fucking cool

raymarched blackhole with volumetric accretion disk r3f.maximeheckel.com/blackhole

LFG!

Reminder for those hesitant to copy patterns and products…your customers do not care where it came from. All they care about is whether your product does the job they need it for. If you care more about pride of authorship than you do about your users, you’re ngmi