1) there is no AI bubble, it has revolutionized how we communicate and learn
2) you don't need to buy an expensive GPU for local LLM, any contemporary laptop with enough RAM is sufficient to run an uncensored Gemma-3 fast
Railroads revolutionized transport and yet railway mania was undeniably a bubble. Something can be both very useful and yet also overvalued and overhyped leading to significant malinvestment (sometimes, everyone wins to the detriment of the investors, sometimes just everyone loses out because a huge amount of effort was spent on not useful stuff, usually somewhere in between).
I'm shocked by this perspective, and I'm deep into the LLM game (shipped 7 figure products using LLMs). I don't feel like anything has been revolutionized around communication - I can spot AI generated emails pretty easily (just send the prompt, people). On the learning front I do find LLMs to be more capable search engines for many tasks, so they're helpful absolutely.
I'm an AI fan, but there's clearly a desperate attempt by just about every tech company to integrate AI at the cost of genuinely productive developments in the space, in a manner that one might describe as a "bubble." Microsoft's gotta buy enough GPUs to handle Copilot in Windows Notepad, after all...
Calling it desperate is a subjective assessment. Yes, some strategies are more haphazard than others, but ignoring generative AI currently is the same as ignoring the internet in 1999 or mobile in 2010 (which facebook famously regretted and paid $4+1B to buy instagram and whatsapp in order to catch up)
Yep — some components currently rely on external APIs (e.g. OpenAI, MathPix),
primarily for stability and ease of deployment during early release.
But I’m planning to support fully local inference in the future to eliminate API key dependency.
The local pipeline would include:
• Tesseract or TrOCR for general OCR
• Pix2Struct, Donut, or DocTR for document structure understanding
Destined to become "best seller" if Aldi/Walmart/Ikea could offer 75" with 4K RGB for $150-200. This would be a perfect indoor wall decor in any room, kind of ultimate poster.
A 31.5" color e-ink poster is ~$1700[1]. I do not know what these large panels cost but (possibly due to low yields) eink panel prices seem to scale superlinearly with area. I would expect a 75" panel that is five times larger to cost more than a used car.
Anything can become a "best seller" if you are able to arbitrarily lower the price by multiple orders of magnitude -- $500 luxury cars would fly off the lot but would not be profitable.
Lol you can hardly get a 7.5’’ for that price. I once looked into making a few programmable ePaper “posters” for myself, and noped out of it when I saw the price.
My understanding is that the yields on large eink panels are horrendous (which is why they cost thousands and thousands of dollars). I would definitely buy one at that price though.
I think the message everyone now accepts is: "there is no moat". It is plain stupid to think big models can be magically copy-protected - they are simply arrays of numbers and all components one need to create such arrays are free and well established. This is unlike the whole infrastructure, processes, social connections, hardware and storage, one need say to recreate a service like YouTube or Facebook. Large models are different - you don't need all of that - the future of LLMs is Open Source like Linux.
Taking the whole software/communication/hardware complexity into account, imagine, how many bugs and vulnerabilities may be present in such non-open-source systems? Based on the recent history where we know even 0-day phone exploits exist, how dangerous would it be if a vehicle (or most of them simultaneously) can become being controlled remotely?
Don't put all your eggs into one basket - what are some good free alternatives? I mean comparable "usefulness" including image understanding, voice interface, and general knowledge?
I’ve been loving having mistral-nemo on my laptop. Not comparable for images, voice, etc, however it is very nice when you’re away from the internet. For the cost of a couple gigabytes you get to keep a good amount of info at the ready. Very easy to run models these days, install Ollama and then do `ollama run mistral-nemo`.
Plus mistral-nemo in particular has a large context window, so you can cook up some shell scripts to throw a bunch of context into the buffer before your question. One I use a lot takes the name of a manpage and a question about it, then the LLM has the whole manpage to reference.
What's the difference when using your own anthropic key vs vanilla? Are you hitting limits with Claude in cursor that they key unlocks?
I ask cos I use it very liberally and haven't had any issues that have made me consider adding a key, except when I made it read my whole codebase on every request
If you're using ChatGPT actively, then surely you have heard of Gemini and Grok (no clue of how far Grok gets you nowadays, but Gemini should. Not sure how good the voice interface is).
I wouldn't use Grok. Google is a big company that has to be reasonably objective, but Elon seems to be the sort of guy who would pettily include jabs at people he doesn't like into the data.
Also he's a horrible human for many reasons and I'd prefer not to support him if I can avoid it. (You know it's bad when Google is ethical in comparison.)
According to the Wikipedia article, Google pulled out of Project Maven due to employee protests. Microsoft and Amazon also worked on Project Maven and Wikipedia doesn't mention them pulling out. So I think Google is more anti-Maven than Microsoft and Amazon.
That's his whole point. And that Elon is even more biased and actively purposefully horrible and unethical than that. By validly criticising Google, you're just reinforcing his point.
Didn't you read the part where he wrote "(You know it's bad when Google is ethical in comparison.)"?
Do you know of anyone at Google who overpaid billions of dollars for a popular widely used communication platform just to use it to publicly humiliate, deadname, misgender, and bully their own child in front of millions of people?
And do you actually think he has the self control not to inject his own prejudices into the LLM he made for that very purpose? Of course it's ingested the sewage of content from Twitter, which is FULL of his own jabs against people he doesn't like, including his own child. He gives his own tweets extra weight, so don't you think he does the same with training Grok?
>Elon Musk's transgender daughter, in first interview, says he berated her for being queer as a child.
In an exclusive interview, Vivian Jenna Wilson said her father’s recent statements, including that she is “not a girl,” inspired her to speak out: “I’m not just gonna let that slide.”
"Do you know of anyone at Google who overpaid billions of dollars for a communication platform just to use it to publicly humiliate, deadname, misgender, and bully their own child in front of millions of people?"
It got him the adviser role of the president, which in turn might save and make him billions.
But his main motivation might have been indeed to fight "the woke terror".
It's a delusional world you live in where you think I was making that argument.
So you have no problem with him abusing his kid in public, because child abuse is ok as long as you don't actually murder them?
Edit: So abusing your kid in public isn't evil? You're fine with that, as long as he doesn't kill them? Isn't it also evil for you to defend Musk's child abuse?
Again, you're missing the point that your valid criticisms of Google only reinforce his point that Musk is even worse.
(If you have showdead=true you can see the idiotic hateful comments in this thread from the kind of people Musk inspires by abusing his child in public. Do you agree with decremental?)
You're still purposefully ignoring and refusing to acknowledge my point that your valid criticism actually reinforces his point of how vile, unethical, and evil Elon Musk is.
You're also ignoring the point that another poster, a Google employee, wrote that Google pulled out of Project Maven due to employee protests, but Amazon and Microsoft didn't. And I'd bet you anything that Musk would gladly accept such evil government contracts for the right amount of money. He already does, in fact.
Please answer my question, if you're not afraid to: So abusing your kid in public isn't evil, public humiliation and verbal abuse is fine parenting, but you draw the line at murder?
So do you agree with the [flagged] [dead] comments of other idiotic hateful transphobic homophobic Musk fan-boys in this thread who are parroting and amplifying Musk's abuse against his own child?
Those [flagged] [dead] posts are incontrovertible proof that Musk's public abuse of his own child actually encourages other people to pile on and abuse her too, as well as many many other trans and LGBTQ people. And they don't stop at Musk's daughter, and the don't stop at verbal abuse: they physically assault and even rape trans people, because people like Musk encourage in incite them to hate and assault the same people he does, including but certainly not limited to his own daughter.
Go to your user use page at https://news.ycombinator.com/user?id=llm_trw , then select showdead: true, then come back to this thread and read the [flagged] [dead] comments, then tell me if you agree with them and Musk, and that's whose side you want to support in this debate. Or just don't reply if you're too embarrassed and cowardly to admit it.
Do you really want to use an LLM pre-loaded with Musk's hatred and abuse? And are you actually naive enough to think he wouldn't do that, since he bought Twitter for the express purpose of shoving his opinions down everyone's throat?
Apart from you being a transphobe, there's also Musk's antisemitism, racism, censorship, and many other things, so this doesn't invalidate my main point at all.
You may as well call me a Hubbardphobe for not accepting the nonsense of body thetans and telepathic exorcism. Most people don't believe what you believe.
I don't agree that Musk is a racist. But I do agree with you about the censorship. All his talk of turning Twitter into a free speech platform was a load of hot air, like much of what he says.
I certainly agree with Yawrehto that you're a transphobic bigot. Whitewashing and carrying the water for Musk's public child abuse and malicious parenting, as well as his well documented racism, just goes to show what kind of a sociopath you and Musk really are.
But thanks for serving a purpose of the shining example and incontrovertible proof of exactly what kind of ignorant hateful pathetic people Musk fan-boys really are. You've perfectly and unwittingly illustrated and validated both my point and Yawrehto's point. Thanks for playing.
Boy, the transphobic bigots like you always crawl out from under their rocks as if on cue whenever there's a chance to suck up to Elon Musk and lick his boots.
Is it comparable with OpenAI’s and Anthropic’s models? I have a strong resistance towards using Musk‘s products nowadays, but maybe I should take a look over the fence.
In terms of general purpose usage and image understanding, I think Grok 2 is pretty good and roughly on par with ChatGPT 4o. Grok 2 will also look for online sources similar to ChatGPT search which is nice. I've occasionally had it report irrelevant results when doing so though --- for example when I was looking up the price of a car it once found some random webpage with the price in rupees even though I'm in the US.
For logical reasoning of course chain-of-thought models like the O1 family are better.
reply