andy99

The gguf is up and works, I don’t know if it’s them or unsloth that’s facilitated this but it’s nice because e.g. Inkling still doesn’t appear to have support in llama.cpp which makes it irrelevant to a class of user.

Unfortunately I don’t have enough experience with Qwen 27B to immediately compare, but I do it’s Qwen 3.6 35B A3. It’s much slower obviously but it seems to be way more efficient with its thinking to the point that using it might actually be faster. I find Qwen and some others rehash the same things over and over when thinking without getting anywhere, in mg limited checks here Muse is much better.

> I'm actually going to make a prediction here as well. I think you will soon realize that using LLMs to clarify certain questions or ideas you have will turn out to have frustrations as well.

Much more likely that people will believe themselves to be an expert in a subject after having had a conversation with Claude about it.

> image processing and resampling/resizing C++ code I hadn't touched in 12 years or so.

So an unimportant, personal project. People seem to extrapolate being able to do something cool into being able to do useful work, which is what this whole discussion is about.

The sycophancy is also a concern, it’s not really an impartial teacher, all its training is to suck up and maximize engagement rather than learning. The incentives are wrong.

What’s the best possible outcome from posting on LinkedIn? I personally think it’s a poor place to “build your brand” - it’s mostly your friends or existing professional acquaintances that see your stuff, and whether you’re looking for a job or trying to do business development, those people are rarely your target market. I think the engagement is better because you have a natural “following” of all your friends / colleagues, but it’s meaningless.

This is par for the course for Li, I’ve found it seems to be about activity or making you feel like you’re doing something, without actually getting any value and in reality just scrolling social media.

YMMV of course, maybe there are real people who’ve had great value from posting?

I used to waste lots of time scrolling LinkedIn, I don’t really know why as it was never great but sometimes you learned something interesting.

Now I find it the opposite of addictive because of how bad it is. It’s just AI slop, both the copy and the pictures, there’s virtually no hope of anything engaging. Maybe it’s just me, it feels like AI has broken their algorithm, there’s nothing at all worth showing. Even the stupid business parables are all worse because they’re just Claudeslop.

Anyway what I find most interesting is I recently signed up for twitter for work (to like some posts in a campaign). I hadn’t looked at twitter in years, but found now it’s almost exactly the same as LinkedIn (the “for you”) part, it’s just inane AI engagement farming that is so uninteresting you don’t want to scroll for more.

Anyway, that’s dead internet theory, I think they’re both done (either that or the algorithms know me really well and just recognize my unadmitted love of AI slop about tips for getting cheap airfare or something)

There aren’t any relevant ones. I think it should be an important goal of these projects (unless there is a clear conflicting goal) to make models people actually talk about and use.

There is no question this is true for the Chinese open LLMs. GPT-OSS had a small moment of interest, arguably it was a success as an open model for a while but it’s not relevant now. The early llamas were probably the most successful for their time.

Allenai / olmo was never relevant as far as I can tell. It’s not super helpful, especially as a sovereign government initiative to build an also-ran, they should be going for real relevance.

  By releasing models with open-weights, DOE seeks to galvanize the scientific and AI communities around shared infrastructure for science: enabling new workflows in materials discovery, energy systems, earth systems modeling, fusion, biology, high-energy physics, and beyond.

This only works if people have a good reason to use it.

> people skilled in a particular areas often feel they are intellectually fit in other areas

This I think is a separate phenomenon, maybe Nobel Disease but there might be a more general term, for example that includes celebrities.

It’s obviously real, at least as used in conversation, whether it meets some rigorous definition I’m sure there’s an out, but we’ve all known these people. With vibe coding they’re everywhere. Is this going to be a modern “begging the question” where everyone knows what you mean but someone pipes up that actually the technical meaning is different?

The OPs framing is better. Where does anyone say there’s a requirement to share data as demanded because you used a publicly available resource. These companies have created something world changing that everyone uses now, that you can get for free or for a very small amount of money. They certainly don’t need to indulge anyone’s nosing or whatever this is, the argument is ridiculous.

I didn’t read it, primarily because of the tulip thing in the title which seems anything but a refreshing take.

> For many sedentary adults, eating less could support healthier aging and possibly a longer life

Benchmarks like the pelican thing are about correlation with “how good the model is”. Better models produce better pelicans.

It’s a useful benchmark (aside from being “cute”) because of its simplicity, both in how many output tokens it takes (though I understand some models think a lot now to do it) and how easily one can subjectively judge. It’s this efficient as a benchmark of performance.

Making a long video takes way more tokens, and presumably is a lot tougher to easily compare. swillison has a presentation that’s pelicans from 2023-present (roughly) showing the progression. Imagine “lord of the rings videos from 2026-2029” or whatever, it would take a long time to watch and be harder to judge, and probably just end up being a comparison of screenshots anyway.

TLDR I feel like the post misunderstands the role of the pelican thing though if find it very hard to believe he really doesn’t understand, so maybe I’m missing something.

I can’t stand AI overviews in almost all situations. There’s something both untrustworthy about it and also the idea that something is being shoved down my throat, like someone is trying to tell me how to think. Plus it basically spoils any kind of discovery, I don’t need the internet treating me like a caricature of an executive that needs everything to cut to the heart of the matter, I’m mostly browsing the internet for leisure anyway, and even if I’m not I want to see the nuance myself.

Any “summarize with AI” is clearly one of those solution looking for a problem things, AI can do it so it gets added in whether anyone wants it or not.

They talk about partying all night and then going to work on drugs.

When I read things like this I feel like I’ve lived a sheltered life, and am happy about that.

I feel like the arguments being made about sharing the training data are from people who don’t have any experience training ML models and have never worked with weights. The biggest reason I have seen that people want the data is to complain about some aspect of it (copyright, bias, etc)

I would be very interested to hear someone explain what freedom it is they need the training data to enjoy.

> obsession with precise terms

They’re not really even being precise. The relevant software freedom, from the FSF is []

  The freedom to study how the program works, and change it so it does your computing as you wish (freedom 1). Access to the source code is a precondition for this.

Ported to the model world, this is fulfilled by sharing the weights and implementation. There’s almost nothing that having the training data gets you (other than actually training it). The weights plus a reference implementation let you see all the states to study the behavior, and let you fine tune it to do your bidding (the abliteration etc). The freedom is satisfied.

Some might argue that without the training data you couldn’t do some classes of experiments to see how it works, say leave-one-out retraining. I’d argue things like that are not really about the model but about ML research or the class of models, which while interesting is not a free software pre-requisite.

[] https://www.gnu.org/philosophy/free-sw.html#four-freedoms

Or the only thing they do in life is come to the office and then mix life with work so effectively it’s a combination of hanging out with friends, eating free food, and working, but it all goes into the work bucket when claims are made about how much they work.

Nothing wrong with that per se, I basically did that in grad school. It’s completely different than actually grinding 100 hours a week though. Also tough to be a part of if you have a life or family outside work.

Do you believe it was this edge case he was referring to in making his point?

A lot of people will stop reading or become completely distracted when encountering something so blatantly wrong, and it also indicates the author is willing to say such things. I see some people downplaying it, that’s fine, it’s clearly not helping the author make a case. Even im this discussion, the current top comment and a long thread are solely about that error (granted that’s not too unusual here)

It’s unfortunate - he’s an author, it would have been fine to stick with an authors perspective that LLMs can’t write (which is true), as well as the copyright stuff (which I don’t agree with but he certainly has standing to give an opinion on).

But he’s made the error of trying to come at it from a technical perspective, when he clearly knows nothing about that side of things, which discredits the rest.

I had seen the cigarette. But but not the meat which costs $2,690 for four tuna-can size cans and is a mix of pork and chicken

Thinking of a concept for a restaurant meal along the lines of a calibrated fried spam dinner, that costs ~2k for about a can worth, fried like they do in Hawaii.

Come to think of it, would love to see an “iron chef- calibrated foods” edition.

Why not? If done right it’s a good way to talk about implications for those companies and provide some education like tailscale did here.

We also saw Anthropic post about “our agent escaped too” and while I understand the incident caused them to review, they found something and needed to disclose, the whole thing came across much worse and largely they got mocked or accused of trying to piggyback, so obviously there are good and bad ways.

If there are other providers with interesting takes, I think they would be worth hearing.

Running local LLMs split across machines. Pretty sure the author has some other posts about doing that.

Edit: https://www.jeffgeerling.com/blog/2025/15-tb-vram-on-mac-stu...

That’s just conflating reasoning and “parallel reconstruction” - there is such thing as reasoning, and I agree it’s probably less common in decision making, but it’s fundamental to many tasks where we figure out a solution, like writing an algorithm, math proofs, etc.

Back in the day it was a bit of a cliche to bring up “clever Hans”, the horse that could do math, when talking about machine learning. He couldn’t do math but he read some cues from his handler of pick the write answers, the handler iirc wasn’t in on it.

The point of the story was that classifiers can be right for the wrong reasons and almost inevitably are. At least there’s zero guarantee that the reason for making the prediction matches the human or “real” reason why it’s correct.

LLMs are classifiers, there is absolutely no reason to assume they’re any different, regardless of any reasoning tokens they emit. They do what their handler wants to see, that’s all, and that’s what they’re trained to do.

People often take this as a knock against them. It isn’t, it’s just the reality of neural network classifiers. The results speak for themselves and don’t depend on whether they “actually” reason, but all evidence says they don’t, or at least there’s no special reason why they would.

AI writing is not good. Seeing this post, and another comment on here earlier today to the effect of “we’ve solved prose”, I’m genuinely really confused. I don’t want to be insulting, even if the first place my mind runs is basically people must not read very much or think very hard about what they read. I can’t imagine a world where anyone ingests a material amount of AI writing and things it’s even “acceptable”. But so many people seem completely blind to it. I’m actually horrified by the idea of a world where people are just slinging empty slop back and forth at each other. Probably many of us have already experienced it at work.

So what am I missing? What’s the most charitable explanation for how anyone can think LLMs are good at writing?

Seems they want the narrative to be that “Claude” (their computer program) independently attacked some organizations, ergo LLMs are dangerous etc.

Another framing would be Athropic irresponsibly (vibe?) coded an attack script, and didn’t monitor it as it was pointed to public facing orgs. There are lots of non-AI attacks a large org with a lot of compute and bandwidth could level against others, there are evidently various failures here, but from a responsibility perspective the conclusion isn’t obviously that AI is an outsized danger, it’s that powerful companies should take care when running security research and not just run things unmonitored against the public.

Then maybe just this timing is really unfortunate, I think most people’s first reaction will be that it looks like a “us too” response to the OpenAI/hf thing.