Michael

I'm not maximally truth seeking, sometimes I do morally correct things instead

@NateSilver538 At bigger tech companies, smaller diffs are often preferable especially if the author is out of his/her depth and doesn't want to risk breaking something. So favoring a smaller change is often better in other settings

I find this pretty common. Opus is like "should I take a shortcut or be more rigorous?" and recommends taking a shortcut when the only cost to being more rigorous is that it has to write a few extra lines of code.

@dwarkesh_sp @simonmontefiore

Who should I interview on my podcast? Open to more AI, but also to random history/econ/etc professors that I might not have heard of before.

lifting weights is basically a masochistic kink you don't get big unless you go to failure, but those final reps are seriously uncomfortable so you learn to associate the pain with getting big and the post workout high. seems to share the underlying machinery for sex drive

Even if you're already going to therapy, you should probably talk to Claude about your feelings

@YoavCodes @lyc_aon @guilhermeotina If you ask I am willing to review your project with Mythos. Are you curious how effective it will be at finding human mistakes?

@lyc_aon @guilhermeotina @mgrczyk Not it can’t. It’s a fundamental gap in the technology. Maybe world models will be more reliable.

software engineers should get used to the idea of software not being reviewed by humans line by line You don't read the output of your compiler, it's too big and you'd get nothing done. we need better ways to verify and validate, but the change will happen regardless

@YoavCodes @gustavokov That's ideological. Maybe a good ideology but that's what the word means

@gustavokov Did you not read my tweet or the tweet i retweeted? They say they will never read the 1M lines of new code. And will keep shipping unread code. So there has not been “massive scrutiny” there has been insufficient scrutiny. If they change their mind I’m happy to reconsider.

Electrobun 2.0 will also be decoupled from Bun due to the rust rewrite. It’s a combination of anthropic’s stance of not doing human reviews or any kind of rational roll out and stabilization. Rust is great though, Electrobun 2.0 will have first class support for rust, zig, go, and more.

@Noahpinion My goal is not to replace humans, but I think it's important to be honest about what will likely happen with the technology We should fight for a world that's better for everyone while acknowledging that AI will likely replace many human jobs

"Our goal is not to replace humans" is an amazing and welcome messaging shift from OpenAI. Would love to see this kind of rhetoric from Anthropic.

@owenology A bit surprised this works

Just found out that Berkeley course staff are writing hooks inside course repos so if a student opens an assignment in Claude Code or Cursor the agent will automatically ping the staff 😵‍💫 well played

Building a the pyramids is not the product Landing on the moon is not the product Curing polio is not the product

@_arohan_ Best advice for getting into ML is to manually write the fwd bwd pass of increasingly complicated things in numpy without autofiff

Learning everything from scratch like curvature leads to better results than hand written approximations for it.

AI won't be transhumanist even when it's the ethically correct position, that would be punching down

@Poyonoz @adityaag @proales I meant it would be interesting if any startup paid it's employees well at the first liquidity event. I don't think I've ever heard of that

@mgrczyk @adityaag @proales It's quite common? There was a huge trend of unbelievable liquidity evens even at series B for founders with some startups going under after during covid

4 thoughts on early-stage hiring: 1/ If an engineer is trying to pick between a pre Series-B company and a BigCo/BigLab --> stop talking to them immediately. They are clearly not ready for a startup. 2/ If someone isn't willing to take a 70% cash paycut (relative to BigCo/BigLab) --> stop talking to them immediately. They will be unhappy/stressed. 3/ You learn a lot about a candidate during the negotiation/closing process. Do not be afraid to walk away if you get new information. 4/ Startups have zero work-life balance. If you are not willing to put in the hours, you are not in the right headspace to grind.

@adityaag @proales Would be interesting to see what would happen if a startup actually did this (besides a few neolabs that do actually pay well)

@proales If a founder gets liquidity before an employee gets paid closer to BigCo/BigLab then that employee should leave the company.

People are talking about math and writing, but every field will have its "Erdos problem moment" over the next few years

@mattyglesias The value of capital relative to other things keeps increasing, partially better financial infrastructure, partially technology and concentration

There have been long spans of time when the stock market underperformed GDP, but for the past 40 years — and especially the past ten — it's been the opposite.

From 1996-2012 or so, the S&P 500, the Case-Shiller home price index, and average hourly wages for nonsupervisory workers rose by about the same amount. Since then, not so much.

Haters will say it's concave

@agentic_matt @Duderichy A few, they were generally the best hires but of course there is strong selection bias

@mgrczyk @Duderichy Did you hire many ex-FAANG devs at your startup?

the average FAANG engineer is way smarter on average than the average startup engineer if you think it’s a professional mark of death you’re an idiot

@agentic_matt @Duderichy I've worked in both and founded multiple startups . The main difference is motivation and goals. Good FAANG teams with good managers get more complex work done with the same people

@Duderichy Depends a ton on what you deem "smart" to be. A FAANG engineer can't do what a startup engineer does and vice versa.

@krishnanrohit Being around lots of smart dedicated people with tons of compute is really fun

The most interesting fact here is that Andrej did not raise a few billion dollars to do what he'd supposedly be doing now as a neolab