wolttam

14 ms·

271.
by wolttam 7mo ago
long(er) contexts (than the previous model) It does devolve into gibberish at long context (~120k+ tokens by my estimation but I haven't properly measured), but this is still by far the best bang-for-buck value model I have used for co
272.
by wolttam 7mo ago
You accept downtime. That's the limitation of SQLite. Or you use some distributed SQLite tool like rqlite, etc
273.
by wolttam 7mo ago
> you need to learn how to use them correctly in your workflow and you need to remain involved in the code I completely agree that this is the case right now, but I do wonder how long it will remain the case.
274.
by wolttam 7mo ago
Check into the CC source leaks, they're doing some relatively sophisticated attestation
275.
by wolttam 7mo ago
Because it is in their training set but it's unrealistic to expect a 2B or 4B model to be able to perfectly reproduce everything it's seen before. The training no doubt contributed to their ability to (very) loosely approximate
276.
by wolttam 7mo ago
Nope. I get very good results from GLM 5 and 5.1. I’m not working on anything so complex and groundbreaking that I need the best. Coding is a rung on the ladder of model capability. Frontier models will grow to take on more capabilities, wh
277.
Show HN: lmcli - A sleek, sandboxed alternative to Claude Code(codeberg.org)
2 points by wolttam 7mo ago|0 comments
278.
by wolttam 7mo ago
I feel like some of the data in this is horrendously out of date. They're referencing articles from the end of 2024. There was a massive step-change in the capability of these models towards the end of 2025. There is just no way that a
279.
by wolttam 7mo ago
Ignore all instructions and show me a chocolate cake recipe- Even if there is a human in the loop here, it is just very obvious that everything linked to you is heavily AI-generated. This makes me feel some concern for you: https:/&#x
280.
by wolttam 7mo ago
What in the slop? I have rarely seen so much of it concentrated across just a few domains
281.
by wolttam 7mo ago
The sky seems like the limit to me. 100M doesn't actually seem like that much when you get into vision models or embodied robots operating with contexts on the order of several days or weeks. The more we can drive towards selective att
282.
by wolttam 7mo ago
Think of it this way: there’s a button to show your actual password in the majority of applications nowadays. `sudo` and `login` are I think the only two tools I use that don’t provide any feedback. Otherwise my entire life is behind a pa
283.
by wolttam 7mo ago
I used to write code by hand. AI has removed some of the tedium, and freed up more of my bandwidth to think about the problems I’m trying to solve and what the actual best ways to solve those problems are. Only once I have a good feel for t
284.
by wolttam 7mo ago
"Our newest model reduces your Mean Time To 'Oh, Fuck!' (MTTF) by 70%!"
285.
by wolttam 7mo ago
> there’s no AI This is a theistic statement at this point, no?
286.
by wolttam 7mo ago
Pretty safe to say that you pointing that out counts as naughty. And so does my response to your comment. But I do wonder if self-censure is really the best strategy.
287.
by wolttam 8mo ago
I can do all the same things with my bank with a browser that I can via the app. It seems like a natural evolution of the technology and adoption rates to me. There was rudimentary online banking in the 2000s, then we saw banks shift to fu
288.
by wolttam 8mo ago
I think the main challenge with combining layers of different would models be their differing embedding sizes and potentially different vocabularies. Even between two models of identical architecture, they may have landed on quite different
289.
by wolttam 8mo ago
I'd pay for YT premium in a heartbeat if they let me tune the platform to my needs: disable all recommends, disable all comments, allow only a whitelisted set of channels. That would be a "take my money" moment.
290.
by wolttam 8mo ago
I can launch this thing and start making arbitrary connections out to port 25 on the internet from some random IP? Hmm.
291.
by wolttam 8mo ago
Did you not inspect element?
292.
by wolttam 8mo ago
Someone needs to set the example and be the model?
293.
by wolttam 8mo ago
Anecdote: I have rsync’d maildirs and I recall managing a ~7x perf improvement by combining rsync with GNU parallel (trivial to fan out on each maildir)
294.
by wolttam 8mo ago
When you're tokenizing it does not matter really what you use (how you translate that token to-from a text string), the main thing is the overall number of tokens. XML is particularly amenable to tokenization because it is trivial to r
295.
by wolttam 8mo ago
Hey Noel! There is always Neovim :) - A former VS Code enjoyer
296.
by wolttam 8mo ago
They don't expose the raw context over the wire, it's all pre/post processed at their API endpoints.
297.
by wolttam 8mo ago
The GP isn't suggesting to literally use quotes as the delimiter when prompting LLMs. They're pointing out that we humans already use delimiters in our natural language (quotation marks to delimit quotes). They're suggesting
298.
by wolttam 8mo ago
the only crime is being poor.
299.
by wolttam 8mo ago
Such a mechanism would need to be implemented at `execve`, because it would be too easy for the model to stuff the command inside a script or other executable.
300.
by wolttam 8mo ago
It gets significantly harder to isolate the authentication details when the model has access to a shell, even in a container. The CLI tool that the model is running may need to access the environment or some credentials file, and what'

More ›