Unfortunately HN changed the direct link to the post: https://discuss.haiku-os.org/t/haiku-nvidia-porting-nvidia-d...

Release link: https://github.com/X547/nvidia-haiku/releases/tag/v0.0.1

Do people who contribute to YouTube Shorts get paid a lot? It's a very nasty thing in the current form.

> What I needed was not a better app, but a way to tolerate imperfection without the whole system falling apart.

> Claude did not invent that idea. It executed it.

> Claude handled implementation. I handled taste.

This style of writing always gets me now :)

This style of writing isnt human. Its AI.

^^ These dramatic statements are almost always AI influenced, I seem to always see them in people's emails now as well. "we didnt reinvent the wheel. we are the wheel."

You are absolutely right. It’s not X that’s the give away, it’s A, and B — moreover it’s C that is the clincher.

AI is popularizing a writing style that has been common in advertising for quite some time. For example, Apple uses it a lot. Now everyone can imitate advertising copy.

It's not that it didn't exist before. It's that it wasn't overused.

Heh.

Neurodivergent people tend to write like this too, there was a study about it.

Yeah, I know, I just feel like at this point it's useless to call that out.

We can learn useful rhetorical techniques from AI that can help us clearly communicate. We should separate those babies from the bathwater.

So far, any "useful rhetorical technique" one could've learned from AI has become a dead giveaway of AI slop (lazy writing and lazy thinking).

Seriously: what tool do you want to use that's immediately available to the absolute lowest common denominator "writers" on the Internet?

"It's not X, it's Y" literally makes my stomach churn from seeing so much of it on LinkedIn.

Rhetorical techniques that are so easily identifiable as AI now.

You can also just find a book on writing. I recommend "On Writing Well" by William Zinsser. Dense and quite informative.

And the scripts of most recent YouTube videos, and the dialogue in Stranger Things Season 5 (the last 3 episodes specifically).

Others, please chime in, I want to take a sort of poll here:

I usually grimace at "GPT smell". The lines you quoted stood out to me as well, but I interpreted them as "early career blogger smell". It's similar, but it didn't come off as AI. I think because it avoided grandiose words, and because it reused the exact same phrase format (like a human tic) rather than randomly sampling the format category (like an AI). This is what human text looks like when it's coming from someone who is earnestly trying to have a punchy, intentional writing style, but who has not yet developed an editor's eye (or asked others to proofread), which would help smooth out behaviors that seem additive in isolation but amateur in aggregate.

Did others share the impression that it's a human doing the same classic tricks that AI is trained to copy, or does anything in this category immediately betray AI usage?

This post felt AI-touched to me, but the usage falls on a spectrum. You can write the whole post yourself, have an LLM write the whole post, or - what I suspect is the case here - have the LLM "polish" your first draft.

Many weaker or non-native writers might use AI for that "editor's eye" without realizing that they are being driven to sound identical to every other blog post these days. And while I'm certainly growing tired of constantly reading the same LLM style, it's hard to fault someone for wanting to polish what they publish.

I feel ya but it didn't get me on this one for some reason. But it gets me a lot on Linkedin - due to which I lost control and blasted off a post yesterday.

I think it some kind of value - vibe dynamics that play in making the brain conscious about it being written with AI or otherwise.

Right now it seems like the project is just a thin wrapper over Brave Search. Building a complete search engine is way harder than that. You could look into using https://github.com/MarginaliaSearch/MarginaliaSearch if you want to run a real search index - https://marginalia-search.com/ is powered by it.

Yup, it is pretty much just a better frontend for existing search. I want to build my own index and ranking algorithm in the future, but sadly it's quite resource intensive so it will depend on financial viability a bit in terms of timeframe.

So it's not a search engine, it's a frontend for a search engine.

You can just turn off the AI feature in Brave search so it’s sort of extra pointless.

It’s possibly worth pointing out that the about page doesn’t offer any indicator that this is an actual nonprofit entity from a legal standpoint, so at this point I have to assume it’s just a sole proprietorship that is pinky promising to become a non-profit.

In that sense I’m quite happy “donating” to Kagi to provide a stable and supported product from a company with employees.

That's fair enough. For the record I do intend to apply for a non-profit official entity. I would say it still has a role as opposed to Brave considering the lack of advertising though.

I certainly assume your heart is in the right place. I felt like it was worth warning people to do research before making donations.

Sure, would've made for a chunky post title tho :P

There's tons of these frontends, including SearXNG and proprietary (but very good) Kagi. Kagi are working on their own index; this will be their meat.

I am convinced LLMs are the way forward for searching, with a caveat: what they summarize isn't very relevant (it is overrated). It just gives a (hopefully accurate) semantic context. What matters is the sources it directs to. These are your links normally on top of ypur search query.

I'm genuinely curious why "LLMs are the way forward for searching".

Is it that the results won't be stack ranked lists anymore and instead a conversational output? Personally that's not what I want. I want results that are contextual to my search. If there's a use case for LLMs in search this would be, at least for me, what I'd be looking for. It seems, however, that all of the AI in search results today are not that.

I do pay for Kagi and will continue if the quality of the product continues to offer the quality product that it is today.

openwebsearch.eu wants to build an open index. But despite the funding it is still quite sparse.

How is it better?

A small correction: Opus 4.5 is included in the Pro plan nowadays, but yeah, the usage limits for it on the $20 sub are really, really low.

It's https://theaidigest.org/village which runs different models with computer access, so Opus 4.5 got the idea to send that email.

I get that the water stereotype is funny, but it gets tiring after a while (because it's not actually true).

For what it’s worth: yes, it’s not technically true, but the reason it’s sticking around is because it conveys a deeply felt (and actually true) sentiment that many many people have: the output of generative AI isn’t worth the input.

Well, it more demonstrates that people will quickly latch on to convenient lies that support what they want to be true, yet impede real discussion of the trade offs if they can’t even get the basic facts right.

It’s not a good thing.

I'm not saying it's "good", I'm just saying that it's worth a qualitative consideration of what it _means_ that this incorrect statement is so persistent beyond "not true, STFU"

Urgh, I know that it's a solid explanation but I hate the "it may not be true but it captures a truth that people feel" argument so much!

See also "instagram is spying on you through your microphone". It's not, but I've seen people argue that it's OK for people to believe that because it supports their general (accurate) sentiment that targeted ads are creepy.

If they are using the microphone to target ads, show me the sales pitch that their ad sales people use to get customers to pay more for the benefits of that targeting.

(I have a ton more arguments if that's not convinced enough for you, I collect them here: https://simonwillison.net/tags/microphone-ads-conspiracy/ )

I get your point, but can you point to a sales pitch which included "exploit security flaws in Android to improve tracking"? Probably not, but we know for a fact they did that.

Also, your own blog lists an leak from 2024 about a Facebook partner bragging about this ability[1]. You don't find the claim credible (and you might be right about that, I haven't looked into it), but I find it strange that you are asking for an example that your own website provides?

[1]: https://futurism.com/the-byte/facebook-partner-phones-listen...

That claim is SO not credible that I think serious outlets that report on it non-critically lose credibility by doing so.

Seriously: the entire idea there is that there was a vast global conspiracy to secretly spy on people to target ads which was blown wide open by THIS deck: https://www.documentcloud.org/documents/25051283-cmg-pitch-d...

I have already experienced the benefits of sending this to several family members, and I'm thankful for the hard work you put into laying everything out so clearly

AI most definitely uses more water than a traditional full text search because it is much more computationally expensive.

The water figures are very overestimated, but the principle is true: using a super computer to do simple things uses more electricity, compute and therefore water than doing it in a traditional way.

Yes, and a YouTube video more than a text article. etc. etc.

It's a tool. The main question should be: is it useful? In the case of AI, sometimes yes, sometimes no.

I mean, think of it this way. If I built a web app that took HTTP requests and converted them into a YouTube video, then downloaded and decoded that video in software, and then served the request, you'd say "that's stupid - you're using 10,000x more compute than you need to".

It's a tool, and using the wrong tool for the wrong job is just wasteful. And, usually, overly complicated and frail. So it's only losses.

There's also Nim, it shines even more in compile time and has proper compile time macros :)

The frontend examples, especially the first one, look uncannily similar to what Gemini 3 Pro usually produces. Make of that what you will :)

EDIT: Also checked the chats they shared, and the thinking process is very similar to the raw (not the summarized) Gemini 3 CoT. All the bold sections, numbered lists. It's a very unique CoT style that only Gemini 3 had before today :)

I don't mind if they're distilling frontier models to make them cheaper, and open-sourcing the weights!

Same, although gemini 3 flash already gives a run for the cheaper aspect but a part of me really wants to get open source too because that way if I really want to some day, I can have privacy or get my own hardware to run it

I genuinely hope that gemini 3 flash gets open sourced but I feel like that can actually crash the AI bubble if something like this happens because I genuinely feel like although there are still some issues of vibing with the overall model itself, I find it very competent overall and fast and I genuinely feel like at this point, there might be some placebo effects too but in reality, the model feels really solid.

Like all of western countries (mostly) wouldn't really have a point to compete or incentives if someone open sources the model because then the competition would rather be on providers/ their speeds (like how groq,cerebras have an insane speed)

I had heard that google would allow institutions like universities to self host gemini models or similar so there are chances as to what if the AI bubble actually pops up if gemini models or top tier models accidentally get leaked or similar but I genuinely doubt of it as happening and there are many other ways that the AI bubble will pop.

Models being open weights lets infrastructure providers compete in delivering models as service, fastests and cheapest.

At some point companies should be forced to release the weights after a reasonable time passed since they sold the service for the first time. Maybe after 3 years or so.

It would be great for competition and security research.

Yeah, I think it sometimes even repeats Gemini's injected platform instructions. It's pretty curious because a) Gemini uses something closer to the "chain of draft" and never repeats them in full naturally, only the relevant part, and b) these instructions don't seem to have any effect in GLM, it repeats them in the CoT but never follows them. Which is a real problem with any CoT trained through RL (the meaning diverges from the natural language due to reward hacking). Is it possible they used is in the initial SFT pass to improve the CoT readability?

How is the raw Gemini 3 CoT accessed? Isn't it hidden?

There are tricks on the API to get access to the raw Gemini 3 CoT, it's extremely easy compared to getting CoT of GPT-5 (very, very hard).

What are you referring to? I see the 'reasoning' in OpenRouter for GPT-5.2, I was under the impression that is the CoT.

Yes, that's exactly what I'm referring to. When you're using the direct Gemini API (AI Studio/Vertex), with specific tricks you can get the raw reasoning/CoT output of the model, not the summary.

in antigravity gemini sometimes inserts its CoT directly into code comments lol

From a tweet: https://x.com/i/status/2001821298109120856

> can someone help folks at Mistral find more weak baselines to add here? since they can't stomach comparing with SoTA....

> (in case y'all wanna fix it: Chandra, dots.ocr, olmOCR, MinerU, Monkey OCR, and PaddleOCR are a good start)

I've worked on document extraction a lot and while the tweet is too flippant for my taste, it's not wrong. Mistral is comparing itself to non-VLM computer vision services. While not necessarily what everyone needs, they are a very different beasts compared to VLM based extraction because it gives you precise bounding boxes, usually at the cost of larger "document understanding".

Its failure mode are also vastly different. VLM-based extraction can misread entire sentences or miss entire paragraphs. Sonnet 3 had that issue. Computer vision models instead will make in-word typos.

Why not use both? I just built a pipeline for document data extraction that uses PaddleOCR, then Gemini 3 to check + fix errors. It gets close to 99.9% on extraction from financial statements finally on par with humans.

I did the opposite. Tesseract to get bboxes, words, and chars and then mistral on the clips with some reasonable reflow to preserve geometry. Paddle wasn’t working on my local machine (until I found RapidOCR). Surya was also very good but because you can’t really tweak any knobs, when it failed it just kinda failed. But Surya > Rapid w/ Paddle > DocTr > Tesseract while the latter gave me the most granularity when I needed it.

Edit: Gemini 2.0 was good enough for VLM cleanup, and now 2.5 or above with structured output make reconstruction even easier.

This is The Way. Remember AI doesn't have to replace existing solutions but can tactfully supplement it.

Is DeepSeek's not VLM?

after clicking on your link I browsed twitter for a minute and damn that place has become weird (or maybe it always was?)

As someone who has been on Twitter since 2007, it’s radically changed in the last few years to the point of being unrecognizable.

Also, do you know if their benchmarks are available?

In their website, the benchmarks say “Multilingual (Chinese), Multilingual (East-asian), Multilingual (Eastern europe), Multilingual (English), Multilingual (Western europe), Forms, Handwritten, etc.” However, there’s no reference to the benchmark data.

I'd want to see a comparison with Qwen 3 VL 235B-A22B, which is IME significantly better than MinerU.

On the OP link, they compare themselves to the capabilities of leaderboard AI's and beat them.