ch_sm

to be fair, nobody in 1990 decided between NeXT and Linux, because (A) Linux didn’t exist yet and (B) they were / would have been completely different markets and totally different capabilities.

I can imagine that the new vapour chamber they’re introducing helps or is necessary to run this chip at high speeds.

I don’t know if i want to expose my child to auto-censorship (or online gaming anyway).

unpopular opinion apparently, but i don‘t understand the finder hate. as a real-life, non-software-developer user, i couldn‘t care less about library folders, root folders, or a unix-tradition file system at all. i want my folders with my documents, pictures, etc. to me that‘s far more user-friendly and comprehensible than what explorer exposes to normal people.

I’m not an expert, but my understanding is that MTPs are smaller LLMs fine-tuned to "mimic" / predict a specific model’s response. It’s possible that the MTP you’re using isn’t trained well enough on Qwen 3.8. What accept rate are you getting?

It depends on your use case, but the smaller Gemma 4 models or qwen3.6:9b would probably run OK on that. I recommend trying it, even just for fun. It‘s easy with omlx.

I‘ve been considering applying at WikiMedia for a while now, but [my country] is not in the listed countries. Is there a chance you‘ll expand this list in the future?

beautiful!

great tip, thanks!

To be fair, that is what it says on their website.

that is a very fair point i haven‘t seen being made thus far. it took a long time for gemma4 and qwen 3.6 to reach the speed and reliability that they have now, and many (including myself) made assumptions about their viability on day one. we ought to give ourselves and the foss runtime maintainers, and quants more time with new models. it‘s not a quick and easy process to optimize at this point.

prooof-read by Qwen

> I'm unable to find a local model that comes close to the effectiveness of GPT models in Codex

I agree with you there, the local models are not as capable as frontier remote LLMs. If you‘re used to letting fable run for an hour to do a novel or complex task, you‘re (probably?) not going to be happy with local LLMs. But, to me, for my daily work, they‘re still usable, and often surprisingly capable. YMMV

> I tried running a smaller model locally, and it's not usable for me.

If you have the hardware, a MacBook Pro for Qwen 3.6 35B A3B and Gemma 4 26B A4B for example, they are absolutely usable, both in terms of speed and quality. Anecdotally, I can use Qwen for day-to-day coding tasks in TS and Go, without hickups.

Right, my bad; I know they claim to not store _all_ prompts and responses, that was a bit cynical/hyperbolical of me. However, regulation could force them to do it — with all the downsides that come with that.

Your privacy argument, on the other hand, makes total sense. Not even security measures like homomorphic encryption or locality-sensitive hashing could fix that basic issue.

here‘s what i don‘t get about this whole discussion. AI companies already store all prompts and responses for future training.

just make an API that returns the string distance between a previously generated paragraph and the query?

that would sidestep this whole problem class.

regulators could even specify how that has to work.

what am i missing?

yes. that‘s the first half of the cause (the other being that he‘s a shameless scammer).

While these arguments are valid, you’re missing the forrest for the trees. Renewable energy has better long-term ROI because it doesn’t require constant purchase of fuel. Renewable energy plants also produce far, far less CO2 that goes into the atmosphere and endangers our biosphere (yours and mine).

> bought it off CraigsList for $10, handed off in Times Square

haha that‘s a nice nostalgic image. I remember installing it on a friend’s laptop with my DVD in a café, to „fix her macbook“.

you can try ollama, omlx or llama.cpp for instance to download a model and get an inference server running locally. They expose „open ai compatible“ endpoints, so you can configure almost any harness to use them.

oh, it‘s the CEO of a well known AI company educating us all – and all he wants is us to hear his hunch on open weight models?! Amazing, please help us understand the situation a bit better, thanks

ah finally. Compiz for react.

In my experience, yes. A bit more reliable than gemma for me. I mostly use A3B (35B, mix of experts) though, because it‘s faster, and in the same ballpark intelligence wise as the dense 27B, so it’s the sweetspot for me. I want to try cohere‘s mini code model next, but worried the runtimes aren‘t optimized for that yet.

i see! would you mind sharing your model string?

WHAT. that‘s amazing. thanks so much for sharing, i‘ve been looking for ways to speed a3b up for days. It‘s 11pm here but i‘ll try this right now

Nice, I tried it too with oMLX — agreed, it seems very capable for coding! Was slightly underwhelmed by performance though. I got about ~24 t/s on the ternary version on my M2 Max 64. That’s quite a bit slower than Qwen A3B 35B (4 bit unsloth). How was perf for you?

that‘s … not really how it works. Agents get your code into their context via tool calls, not by uploading the entire file to a server „where the weights live and thus the code has to be too“. Small but crucial difference. Aside from that: LLM providers have to be the ones that facilitate your privacy and security by default. That‘s not on the user. They‘re tool providers and they can‘t compromise you and your org by default.

the point the article makes is good (albeit not new). the style sounds very LLM to me.

>The way royalties get assigned is based on a percentage of your listening versus your monthly payment. For example, spend an entire month listening to Taylor Swift’s new album, she gets the entire royalty share. But if you listen to the album 100 times but then listen to lofi beats 900 times, Taylor only gets 10%.

Not sure about Tidal, but that’s not how it works on Spotify. Instead, artists get paid per stream — about 0.003$. Anything from you the pool of your monthly payment allocated for royalties that you did not use gets distributed proportionally between the artists with the most streams. So say:

- I pay 10$ a month - 5$ of that is reserved for royalties - I listen to 330 songs a month (~ 1$) - 4$ remain in the pot and are paid to the most popular artists