seizethecheese

5 ms·

61.
by seizethecheese 2mo ago
This growth is surely good for github. If another company becomes the "github for AI agents", they'll lose not just their business for AI but also human coders. (Sure, maybe there will be a human coder only github, but it wil
62.
by seizethecheese 2mo ago
Impressive for whom? It's impressive for the service to have such growth at that scale, the code being slop is somewhat irrelevant. Your comment just seems like mood affiliation (AI should be dismissed, growth was from AI, therefore gr
63.
by seizethecheese 2mo ago
Doesn’t work for wsj but not dead
64.
by seizethecheese 2mo ago
Sure, and that would be a reasonable thing to reply to the main comment here, but the actual comment was just muddying the waters and a total non sequitur, as written.
65.
by seizethecheese 2mo ago
Another strange comment, this time from an account made 15 hours ago. My entire point was contextual and not in isolation!
66.
by seizethecheese 2mo ago
Strange comment. The comment you're replying to is pointing out that the CIA bought computers for its own use. Am I confused and they also "provided funding" other than buying computers?
67.
by seizethecheese 2mo ago
Father of three and startup founder here. In college, I burned out on normal coursework and dropped out to teach myself how to code instead. To support myself, I worked in restaurants. I found that working as a dishwasher was quite compleme
68.
by seizethecheese 2mo ago
I don’t think so for two reasons: 1. They still have extreme revenue growth (and despite what HN says, good gross margins supposedly). 2. People will pay a premium for the best model. So far, we see catchup but no leapfrogging.
69.
by seizethecheese 2mo ago
HN commenters shouldn't have to be Sherlock Holmes! Extraordinary claims should be supported by evidence. You could have easily posted the seed, the 50 questions, and the results for claude and graft on each, but didn't. I also se
70.
by seizethecheese 2mo ago
Wow, this is really weird. I looked at the github and see you published benchmark updates in these sizes of N: 9 → 20 → 36 → 50 These are super arbitrary numbers of benchmarks to run (9, 11, 16, 14). I see a few explanations for this: >
71.
by seizethecheese 2mo ago
This looks cool and the mechanism looks plausible. I found the experience of trying to understand whether the claims here are legit to be aggravating. First of all, the whole readme section about benchmarks appears to be Claude/Codex w
72.
by seizethecheese 2mo ago
Mentioned where? What do you mean it wouldn't preserve quality? I demonstrate in the comments here how it would on a saturated benchmark.
73.
by seizethecheese 2mo ago
Okay, that makes more sense, but then this post is very confusing. Why mention the router at all? Or does your live system have a router and the eval doesn't? In which case why not publish an eval with the router?
74.
by seizethecheese 2mo ago
Think of it this way: you are at an enterprise business. You have a workflow implemented a year ago that is working just fine. Swapping out the model for a new one changes behavior in unpredictable ways. Eventually, you'll do it once c
75.
by seizethecheese 2mo ago
I did read "Full results and methodology". It doesn't seem to show which agents you are routing to. Am I missing something? And how do those agents score on SWE-Bench Verified?
76.
by seizethecheese 2mo ago
Maybe the business model is to break even on bleeding edge models while making money on the long tail of usage once systems are tuned for a specific model and running in production.
77.
by seizethecheese 2mo ago
The same thing holds for speed. I could build a system that speeds up Fable on GPQA Diamond ~50%, while improving score, by literally randomly selecting between Fable and Gemini 3.7 Flash. (Solve time for Flash is 0.1min and 0.8min for Fabl
78.
by seizethecheese 2mo ago
This is a promising direction! Unfortunately, I think the benchmark result here is essentially meaningless. I recently discovered this same lesson the hard way. I was trying to get a multi-agent system I was building to improve upon GPQA D
79.
by seizethecheese 2mo ago
Same for the environment and construction. And there’s a correlate for rhetoric too, such as comparing people to Hitler or Stalin.
80.
by seizethecheese 2mo ago
I asked my barista what they think about Qwen 3.8 Max…
81.
by seizethecheese 2mo ago
Which reply? Because both seem like nothing a bot would say honestly.
82.
by seizethecheese 2mo ago
I think you're wrong. To gut-check, I asked a panel of AI models "What platfroms should be considered tastemaker or trendsetter for adopting new AI models". Here's the part of the answer relevant to this discussion: >
83.
by seizethecheese 2mo ago
While composing a reply to a comment throwing tons of shade on American AI, I took some time to check out the commenter’s HN profile. Their comment history was about 50% such comments. Their submission history started with an article about
84.
by seizethecheese 2mo ago
I don’t think business really wants interest rates that high. Without conspiracy theories, a simpler explanation is that this is simply democracy in action. People don’t want the spending reigned in, and they don’t want their taxes raised.
85.
by seizethecheese 2mo ago
It could save billions by raising confidence in predictions. Bad predictions have a “boy who cried wolf” aspect.
86.
by seizethecheese 2mo ago
I'd prefer congress to be asking questions (this is all they are doing so far, based on the article) before doing any legislating.
87.
by seizethecheese 2mo ago
Dad was a real newspaper journalist pre-internet and he used to bemoan the headlines the copy editors would come up with.
88.
by seizethecheese 2mo ago
Latency isn't changing the results for the coding index or arena ELO, but neither of those take latency or throughput into account, so we added those to our leaderboard as score components. Latency and throughput matter a ton as a user
89.
by seizethecheese 2mo ago
Opus is still first in Intelligence Index followed by Fable, GPT 5.6, Kimi K3 then Qwen 3.8 max. https://artificialanalysis.ai/#intelligence Our leaderboard combines Arena ELO, AA Intelligence index, latency and speed and g
90.
by seizethecheese 2mo ago
I ran a hardware startup for about a decade. We ended up doing all PCB assembly in China. We did try to find this in the USA, but it was indeed too slow and expensive. The bottleneck in assembling a PCB is usually component sourcing. My pro

More ›