eli

5 ms·

1.
by eli 5d ago
There’s a surprising amount of variety in between implementations of “OpenAI” endpoints.
2.
by eli 7d ago
You can test it now on the official deepseek api. Just set your model to deepseek-v4.1-flash-expires-on-0910 It’s good and very fast. (Note that the deepseek API trains on your data)
3.
by eli 9d ago
Nah it’s fine. They committed to international devices being unlocked, and the firmware lock is trivial to bypass anyway.
4.
by eli 10d ago
I can believe a lot of people said that. But I'm not sure that means it's a true prediction of consumer behavior.
5.
by eli 11d ago
Easy, Rambo. How about we start with any punishment.
6.
by eli 12d ago
Because he posted about it. It was an incredibly racist post claiming “gypsies” are like invading wolves and that something more drastic must be done to get rid of them before they kill all the “sheep” in Copenhagen.
7.
by eli 12d ago
It's actually a standard term for the person who plays this role during incident response https://www.pagerduty.com/resources/incident-management-resp...
8.
by eli 12d ago
Hours after they alerted Meta that similar activity was happening from their corporate network?
9.
by eli 12d ago
I get what you're saying and it's concerning how much power these big labs have amassed and how little transparency there is in what they do with it... But I doubt this a major factor in the trend. I just don't think it'
10.
by eli 13d ago
Why? Seems like benchmarks that closely mirror the tasks you'd want an LLM to help with would be a lot more useful than some general intelligence benchmark.
11.
by eli 13d ago
The session had a 91.4% cache hit rate. They just give zero discount.
12.
by eli 13d ago
I just did a little anecdotal test. Had pi + cerebras review a recent commit and asked a few quick followups on it. Worked great. The Cerebras session cost me $1.60 and took a total of 5.1 mins. I did get a few brief 429 rate limit errors i
13.
by eli 13d ago
If you read the reasoning trace for Qwen 3.8, it does a whole lot of "uh" and "But, wait..." too
14.
by eli 13d ago
Strongest model that they host on the public endpoint. They do a super fast version of GPT 5.6 Sol for OpenAI and have bigger open models on dedicated endpoints.
15.
by eli 15d ago
OK, clear enough. But that's an unusual definition of "build" that I don't think you should necessarily expect others to share. The first entry for "build" in the OED is: To construct, put up, erect (a house o
16.
by eli 15d ago
The construction manager can’t say they built a skyscraper? In that only for someone who personally laid every brick and ran every wire?
17.
by eli 17d ago
Only to me, the person with access to the session. It implies LLM generated to anyone looking at the commit.
18.
by eli 17d ago
Seems plausible that the feature was only added (quietly and by default) for marketing reasons
19.
by eli 17d ago
It’s a default that changed silently during an automatic update. Also I frequently use an LLM to commit work that I have written. It is just misleading in that case.
20.
by eli 17d ago
Openrouter tracks what apps are using the model and the top ones for hy4 are all different coding harnesses. I guess it could be fake but seems more likely people are just trying it out. Hy3 was a very strong and underrated model.
21.
by eli 17d ago
What if that isn’t the most important part
22.
by eli 21d ago
I mean, you can try it for free.
23.
by eli 22d ago
I have been working on a personal benchmark suite to test new models and ironically one thing all the models are bad at is writing new benchmark tasks. I guess it’s the different layers of abstraction between the task and how it’s evaluated
24.
by eli 22d ago
But it's an error, not a response.
25.
by eli 26d ago
https://www.liberalcurrents.com/the-reconstruction-papers/ is explicitly aiming to be a Project 2028 book for the left.
26.
by eli 29d ago
I think they run whatever models they get paid to run. But mostly from enterprise. They are clearly not interested in consumer dollars.
27.
by eli 1mo ago
Yes - unusually (uniquely?) bad for an official API. Which is a real bummer because it’s otherwise solid with excellent caching.
28.
by eli 1mo ago
The philosophy with Pi is it is minimal (but functional) out of the box and easily extensible. I'm not familiar with that feature but I would not at all be surprised if someone already coded a Pi extension that does it.
29.
by eli 1mo ago
They did release the weights and there are many providers https://openrouter.ai/deepseek/deepseek-v4-flash-0731#provid... Deepseek's official API has a pretty bad privacy policy so I would assume businesses avoid
30.
by eli 1mo ago
Isn't it a good thing to know about price hikes in advance? If I were building a product around it, I would certainly care.

More ›