eli

6 ms·

31.
by eli 1mo ago
Anthropic the other day announced future models "launched in the EU" will include watermarking https://support.claude.com/en/articles/16266773-how-claude-m... I was assuming it was something like SynthID
32.
by eli 1mo ago
Luna has one vendor and they can change the price any time. Deepseek is open and has dozens of vendors competing to serve it.
33.
by eli 1mo ago
I wondered if the promised change in pricing is actually going to be deepseek bringing up their cached costs. They're extremely inexpensive.
34.
by eli 1mo ago
The Deepseek official API is good with excellent caching. But their privacy policy is unusually bad - they can train off your prompts and completions.
35.
by eli 1mo ago
Opus 5 medium to Opus 5 max is only 3 points, if that puts it in context
36.
by eli 1mo ago
Official pricing only kinda matters for an open weight model, no?
37.
by eli 1mo ago
Fable's guardrails would never let it do something like Cybergym so at least for that one it's measuring Opus 5
38.
by eli 1mo ago
The top 4 models on that site are all variants of Gemini Flash? That does not match my experience at all.
39.
by eli 1mo ago
I've seen ones that are configurable to pick a trade off point between lower cost (cache stickiness) and routing performance (best model for that turn). But yeah I'm skeptical all this overhead is worth it.
40.
by eli 1mo ago
It’s injected into an outbound api call, not into an env var the agent can read.
41.
by eli 1mo ago
You can run the agent in the gondolin sandbox if you wish. Their example implementation with pi uses a pi extension so that pi runs on the host but the read/write/bash/etc tools run in the guest. Doesn’t have to be that way t
42.
by eli 1mo ago
That incident was with a model that had the guardrails disabled. Still obviously you should run all untrusted code in a sandbox, but extreme actions like that would be very unusual with the model that shipped.
43.
by eli 1mo ago
I mean…yeah…it’s a bug.
44.
by eli 1mo ago
Qwen 3.8 Max is very strong at troubleshooting and code review.
45.
by eli 1mo ago
I assume/hope this is about prices going up for the next release of Pro
46.
by eli 1mo ago
And all that just to allow internet access for npm and pypi? If you've got the bandwidth and disk space, it's very easy to make an offline mirror of both.
47.
by eli 1mo ago
That's got a significant selection bias. Claude and ChatGPT and Gemini and other subs do not go through openrouter.
48.
by eli 1mo ago
I use https://pi.dev/ which works fine out of the box but is fairly minimal and intended to be customized. There are many extensions. OpenCode or oh-my-pi might make more sense if you just want a batteries-included agent. Y
49.
by eli 1mo ago
It's not exactly what you asked for, but the public transit and multimodal focused apps are generally much better at this stuff. I like CityMapper and Transit. Neither is good for turn-by-turn directions, but both are good for "wh
50.
by eli 1mo ago
It's not enough that it's better? Many providers will host it and will compete on price. It also can't easily be taken away because one company (or one government) decides they don't want it around any more. People can f
51.
by eli 1mo ago
I believe it. It's extremely good at troubleshooting. I gave Qwen and Kimi K3 the same annoying, complicated, intermittent bug to track down. Kimi did a bit better in understanding the existing code, but Qwen built some diagnostic tool
52.
by eli 1mo ago
The amount of abuse of free VMs must be absolutely immense even with credit card verification. I’m not sure there needs to be additional secret motives.
53.
by eli 1mo ago
They're usually leased. So yeah they probably owe Flock new ones.
54.
by eli 1mo ago
They bill out the ass to lease them
55.
by eli 1mo ago
I'm not familiar with North Dakota in particular, but at the federal level and in some states ALL works created by the government are exempt from copyright. The public paid for it, so the public owns it.
56.
by eli 1mo ago
According to a few tasks from my little personal coding benchmark it's very good at coding and kinda bad at web design. (Also excellent at "draw me a picture" one-shot prompts, for whatever that's worth) On a sneaky on
57.
by eli 1mo ago
To be fair, "people with no skills inflating their own value" is what LinkedIn has always been like. But I guess LLMs are uniquely well positioned for that task.
58.
by eli 2mo ago
These are the API rates. They don’t train on prompts from the API. In what sense are you the product?
59.
by eli 2mo ago
I mean, yeah. Alternatively I’ve seen the approach of starter with the cheaper/faster model but give it a tool to handoff or talk to a strong model if it gets stuck.
60.
by eli 2mo ago
This is the improved iOS parental controls. When they shipped it (and for years afterward) it simply did not work https://www.wsj.com/tech/personal-tech/apples-parental-contr... It really is surprising given App

More ›