j_bum

Fair point, wasn’t considering it from this angle.

Do you use Claude code on your machine? That seems mostly vibe coded

Oh my god, my two worlds just had an insane collision.

I learned speed cubing from badmefisto when I was in middle school, ~16yr ago (today my ao100 is ~15s).

I never knew it was Karpathy. What an insane knowledge drop. Thanks for sharing!

Curious what you mean by killing it? Products? Model quality?

Yes, the term is “sacrificed” is unfortunately used today in many research labs in the US.

My advisor had a *strict* policy against people using that terminology in the lab and in his department.

“We sometimes euthanize or kill animals in a part of the difficult process of research. We do not sacrifice them - we are not making offerings to some deity. We are conducting research, not participating in a religion.”

“Sacrifice” is a euphemism that serves only to disconnect the ethical and emotional burden of killing animals for research [0].

And I deeply sympathize with your ethical and emotional guilt. The research I did in my PhD contributed to the foundational knowledge in my field, but not without severe and serious tolls. The way people become normalized to euthanasia in research environments is scary.

[0] https://openworks.mdanderson.org/cgi/viewcontent.cgi?params=...

Would you be willing to pitch why you like jj and find it useful? How were you using git?

I use git mostly on single-dev projects, with branched development patterns.

Would someone like me with a simple git use-case find JJ enjoyable?

GP said “can no longer exist”.

Listing a few challenges of a phoneless life doesn’t negate the point that one can still exist without a phone, even in 2026.

This was very touching, thanks for sharing. Sorry for your loss.

“Has my brain settled” I feel this.

I started meditating recently (~10mins per day) and have found it to be surprisingly effective. It’s a combination of body scanning & mindfulness meditation.

It doesn’t always help, but tends to.

This video gives a great overview of the history of the acceleration:

https://youtu.be/glWvwvhZkQ8?si=-HGtfd_KHYfatEQ

Although it’s focused on Ilya, some great history is covered.

Yep I bewilders me when I see instructions like this. Go bad and edit your previous message if you didn’t et what you want!

I think this is a direct result of OpenAI and Anthropic humanizing these models too much.

I want C-3PO by my side helping me work, not a machine acting emotional.

But that’s what they’ve given us, and now a huge fraction of the username treats these tools like a human.

But how do you know they’re not lying to you? What are your benchmarks for this? Experience? Anecdote? Data?

And I’m asking you in good faith - not trying to argue.

I’m thinking about these types of questions on a daily basis, and I love to see others thinking about them too.

Your display and thinking summery settings aren’t working for me (v2.1.112 on macOS). Any advice?

Claude code hides thoughts - which used to be how I’d stay engaged. Now they’ve got it all hidden.

Otherwise, I love your advice. Thanks.

It’s ok for the mind to wander a bit, and we are lucky to be in a time during engineering where the ability to think has become a priority.

I agree with almost everything you’re saying.

I really love the “brief moments” part too. I believe growth (maturation, skill improvement, etc.) happens in discrete moments that is then reinforced by practice (or lost by the lack thereof).

> stuck at some ceiling… next step forward

Then they weren’t at their ceiling :) they were at a local minimum of optimization. And getting out of a local minimum is insanely rewarding, regardless of the skill.

And maybe you’re right - maybe there is no hard ceiling. But there are learning rates and diminishing returns involved. It might take me 10x longer to get from the 95% percentile than 96% percentile, and then 100x to get to 96.5%. (Obviously percentiles are quite abstract for art). Maybe we can always improve given the right practice and guidance, even if the improvement is marginal. I just define the marginal returns area as a “ceiling”.

But… all of this can co-exist with “natural” baselines of talent.

I hate to admit that I will never be as “good” of a pianist as Martha Argerich… but I just won’t be, no matter how much I practice. I also will never be able to run as fast as Noah Lyles, no matter how much I train.

And that’s ok.

I think there is a tendency to fight back against the idea of natural talent/skill because this idea can diminish the extreme hard work it takes to hone and develop talent.

Noah Lyles could outrun 99% of us without training, but he had to dedicate his life to beat the last 1%. Most people probably don’t realize/understand how much dedication it takes to climb that final mountain.

But “inherent” talent is still a real thing. If we knew “why”.

My final thought is that high baselines can often be counterproductive for growth. Anyway, thanks for chatting about this with me.

Agreed. I actually have thought those were “waiting to get a response from the API” rather than “the model is still thinking” messages

Of course. I agree with you. You will only make it in that field if there’s literally nothing else you can imagine doing.

But there’s still a starting point and ceiling in intrinsic ability.

I’m curious why you think there’s no difference in inherent abilities? Some people are more “naturally” skilled at some things than others, at a baseline. You can look to see examples of this everywhere.

That doesn’t negate the dedication/sweat/tears of people who have high skill, but many of those people also started several miles ahead.

As someone with a BM in piano performance - I’m of the strong opinion that there’s a natural baseline and ceiling. Practice can only get you so far.

As someone who finds a huge amount of enjoyment in developing using Opus 4.6 in Claude code, I’d love to know what other harnesses people use that deliver the same experience as CC. CC is a vibe-coded mess, but it works very well for me.

I do a lot of work in R and find codex (5.4 & 5.3-codex) just totally drop the ball with R. Anthropic’s models are far better with R, so I use them.

But I do wonder how much the harness affects performance.

Would GPT-5.3-Codex perform just as well if it was plugged into CC?

Some feedback:

1. The main page asks for an email to be notified when the hoodie is available to buy, but I can add the goodie to my shopping cart and proceed to check out 2. The product page mentions a 6’ model but there is no model in the images 3. The check out page says “there are no payment options, please contact us”

What self-hosted sync approach are you using?

I’m thinking of doing the same thing, as I want everything on my home NAS and not iCloud/Google Drive

Thanks for sharing your article - very well written.

I am stunned to see that LoC risk assessment.

I kept wondering to myself over the past week, “will this be the last USA-supported human space travel if these astronauts don’t survive?”

I’d have a hard time imagining the general public would support any future missions if they hadn’t survived.

These astronauts are some elite humans. My respect for them is even greater now that I’ve seen the risk quantified.

+1, I’d love to see the app.

Incredible effort… thanks for sharing this!

I’d love to learn more about the technical challenges. For example, how do they handle buildings that aren’t perfectly aligned to the cardinal axes?

Update 13hr later: I was able to claim my credit.

Yes, I did have it enabled. Though the terms say you "must have it enabled", not "must not have had it enabled". Perhaps there is a roll-out difference.

I can “buy extra usage” for “up to 30% off” right now, but no free credits to be claimed on profile dashboard at the moment.

I’m not sure if the analogy is yours, but the scribe note really struck a chord with me.

I’m not a professionally trained SWE (I’m a scientist who does engineering work). LLMs have really accelerated my ability to build, ideate, and understand systems in a way that I could only loosely gain from sometimes grumpy but mostly kind senior engineers in overcrowded chat rooms.

The legality of all of this is dubious, though, per the parent. I GPL licensed my FOSS scientific software because I wanted it to help advance biomedical research. Not because I wanted it to help a big corp get rich.

But then again, maybe code like mine is what is holding these models back lol.

Have you ever talked to him about this?

Could you point me to some resources to learn about harnesses? I’d love to hear an example of a use case you’re thinking of.