defipunk.eth 🦇🔊

@GlitchPoool @johnrushx Have you tried it? Getting spammed with the most mundane commits isn't anything to marvel at...

@johnrushx the hockey stick on that Claude Code commits chart is pretty wild ngl. 4% → 20% of all commits in 12 months is an insane rate. the question is what actually counts as 'authored by claude'

This equals 40,000 full-time software developers working full time. End of 2026: 200,000 developers. 2027: just Claude Code alone will be adding as much code as 1,000,000 full-time human developers. 2028: 1B+ Enjoy your last lines of handwritten code. Horses replaced by cars.

@stevibe @asaio87 Before I cancelled it, I got 1 request per day out of the pro plan with opus. 1. I gave up after 7 days of that.

@asaio87 No, definitely has limit. It is 20x the Pro plan.

Is the Claude max plan at $200/mo really unlimited ? I mean can you code non stop for a month with it ?

@alexkehr You work on very different problems...

hot take: any software engineer who is not already at close to 100% ai coding output should be cut within the next 3 months not only are they inefficient, but they’ve also shown a lack of curiosity and willingness to learn

@thefinnmckenty @GergelyOrosz And all new projects will have the same "style" that the AI prefers. Seems like it will solidify its necessity that way as well...

@GergelyOrosz Yep, that sounds right to me. We have a somewhat niche, poorly documented product at the heart of our stack and AI coding just doesn’t really work with it at all (unfortunately).

Interesting: I am hearing both inside Meta and inside Google AI coding models simply don't work nearly as well as they do outside: Models work great on greenfield projects and when using standard tooling. Both Meta and Google have monoliths and non-standard tooling!

@GergelyOrosz Yeah, Google is back at using kLOC for productivity. Seeing how my colleagues use AI, that's about right.

Dang, 300K+ lines in prod! Inside Google. You heard it here: x.com/rmedranollamas…

I wonder who's responsible @google for pricing gemini-cli... 1000 QPD for Gemini 3 every day is kinda nice, when it's normally just 20, and 5 per minute if you use the API vs `gemini -p $prompt`

@eshear Can't tell without doxxing, but think vscode with LLM extensions, with both fine tuning on internal sources and rag implementations for most internal data + tools for internal tooling

@DefiPunk By “ai coding tools” what do you mean?

METR’s analysis of this experiment is wildly misleading. The results indicate that people who have ~never used AI tools before are less productive while learning to use the tools, and say ~nothing about experienced AI tool users. Let's take a look at why. x.com/METR_Evals/sta…

@eshear Never touched cursor, but i'm using ai coding tools in nearly every session. Your counter is a bit weak here

“We recruited 16 experienced open-source developers to work on 246 real tasks in their own repositories (avg 22k+ stars, 1M+ lines of code).” So they sound like reasonably experienced software devs. "Developers have a range of experience using AI tools: 93% have prior experience with tools like ChatGPT, but only 44% have experience using Cursor." Uh oh. So they haven't actually used AI coding tools, they've like tried prompting an LLM to write code for them. But that's an entirely different kind of experience, as anyone who has used these tools can tell you. They claim "a range of experience using AI tools", yet only a single developer of their sixteen had more than a single week of experience using Cursor. They make it look like a range by breaking "less than a week" into <1 hr, 1-10hrs, 10-30hrs, and 30-50hrs of experience. Given the long steep learning curve for effectively using these new AI tools well, this division betrays what I hope is just grossly negligent ignorance about that reality, rather than intentional deception. Of course, the one developer who did have more than a week of experience was 20% faster instead of 20% slower. The authors note this fact, but then say “We are underpowered to draw strong conclusions from this analysis” and bury it in a figure’s description in an appendix. If the authors of the paper had made the claim, "We tested experienced developers using AI tools for the first time, and found that at least during the first week they were slower rather than faster" that would have been a modestly interesting finding and true. Alas, that is not the claim they made.

So... the workers died when Israel opened fire on the ambulances, then Israel went and handcuffed the bodies before burying them?

@Aella_Girl What about his breeding fetish and him claiming humans are dying out?

i try not to comment too much, i think there's plenty of serious, correct critiques to make about elon, but it seems pretty unlikely to me that he's a nazi. that's just a inefficient, dumb, unlikely position for him to hold, and i find the supposed evidence for it unconvincing

@Aella_Girl Partially a size difference reason in my case. Sitting next to each other, I can't get in that position comfortably

What percentage of cuddles on this position involves/involved the MAN on the WOMAN's chest/shoulder

Straights: in your last/current relationship cuddling, what percentage of cuddles involved the man in the position with his head on the woman's shoulder/chest? Like below, but with genders reversed. Poll in next tweet

@9mmsmg You see how the Chinese economy is doing, right? And the effect their excessive lockdowns had on them?

There is no one who can convince me that China didn't release a bio weapon on the entire globe intentionally. The entire planet should have put aside our differences and united against China. Instead, we just let them grow more powerful. Chinese nationals heading into the USA to buy all our property cheap after. An absolute menace.

Sacks now just posting full on Russian disinformation and fake claims. 1) The US involvement in Ukraine is both the right thing to do, but also our obligation under the 1994 Budapest Memorandum. 2) There has been no fraud related to Biden or USAID proven in Ukraine. 3) The corruption that has existed was worth at most hundreds of thousands of dollars and resulted in Zelensky firing those caught doing so. 4) The claimed proof of fraud was falsified documents, literally produced in a Russian disinfo campaign, as determined by multiple US intelligence agencies and allies, and published by the DoJ. 5) 75% of all aid sent, directly benefited the US as it consisted of old weapons that were legally required to donate to allies or destroy, and the restocking new ones for the US. 6) It also created tons of new jobs in the US because the bulk of the cash was required to be spent with US munitions companies. 7) Another 15% of allocations included humanitarian food aid, or lines of credit with European entities for buying. So next to no cash was sent. 8) Ukraine did not suffer from this aid, instead they held back, for 3 years, someone who was supposed to be a military super power.

@GergelyOrosz How much time was the second-round screening by the first founder or what was the scope of it? Just linked-in/CV check, short telephone call, or even more?

Actual numbers on what it took a seed-stage startup to hire their first two founding backend engineers (full-remote) It's good to know as a candidate: this is the type of competition you might have. And as a founder: this is how much time investment it can take to make a hire

@brogueboy @GergelyOrosz @marsh_lt Depends on how much goes in private bank accounts right away and how much each country will impact production efficiency for political gain - like your average senator. Which is why "every" spending bill needs to have something for nearly everyone...

@ExplorationOnK7 @DuaneStorey @GergelyOrosz They work great against 50 year old Russian tanks. Next gen tanks will be able to counter drones efficiently.

@DuaneStorey @GergelyOrosz I’m curious if any militaries are learning from the Ukraine invasion. Cheap drones and javelins work great against expensive tanks and ships. Russia’s defenses seemed to hold against the counterattack. After Russia‘s initial gains, they have made some, but not much progress.

@GergelyOrosz @DuaneStorey But turning inward has huge consequences. Military goods lose a large chunk of economies of scale and also resale value (spare parts?), intelligence of the 5 eyes now needs to be collected alone, the dollar may lose its reserve status and as a result crash, ...

@DuaneStorey I personally don’t believe this - the US is and will remain the biggest military superpower I’m pretty sure. It simply is turning more inwards and not interested in playing a military role as broadly as before - at least all signs point to this

@Aella_Girl If it helps, I followed you ages ago because I liked your content and probably you, at least your public persona, and you tickled something in my brain with many posts.

i'm kinda angry that it seems like I'm responsive to the opinion of the masses, and also that the thing that shifts the masses is so trivial.

i've heard ppl who lost a lot of weight talk about some angry cynicism when people start treating them better, even ppl they've known for a long time. I'm having a bit of that now that twitter seems to like me. i've been consistently myself this entire time, what's happening.

@GergelyOrosz @NathanpmYoung There is an even worse incentive system to allow large sponsors to provide specific fine-tuning data.

It the past micropayments have not worked for anything like this. Crypto is an implementation detail that made no difference in the (non-) success of micropayments I personally don't see this path working (I could be wrong). I can see ways to connect expensive subs to LLMs over time in some ways endorsed / allowed / tracked by the source vendor.

Fair chance that Deep Research by ChatGPT, Perplexity, Google will only get worse over time: not because LLMs will get worse, but because valuable data for any decent analysis will NOT be on the open web. Paywalls are coming to lock away quality stuff. Even now I can see it:

@GergelyOrosz If Google (and in particular the now ousted SVP Raghavan) hadn't pushed for short term metrics and profits, they would likely still have an industry leading infrastructure to flag these sources with unrivaled trust data. Quite the fumble to put a Yahoo P Person in that position.