Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Grok3 Launch [video]

x.com

Post reply on HN
Print view View original

521–530 of 1001 posts

Re: Grok3 Launch [video]

#521

Post by visarga »

ben_w wrote: post #507
numpad0 wrote: post #446

Earlier quoted context omitted.

Considering that OpenAI subscription is $200 per month, and "Premium Plus" subscription that includes this thing is only $40 per month, does that mean instantaneous "Elon factor" is now at least -$160 per month per user, or is it supposed to be added up to more than -$240 per month? How would the math change after factoring in that OpenAI isn't even covering entirety of opex with the sub anyway, and/or people finding…

Given how fast-moving the field is, it's very difficult to confidently state how much inference costs. Perhaps he's under-charging, perhaps OpenAI is over-charging, one may be more optimised than the other, but new models come out and change everything in less time than is normally takes for actual costs to become public knowlege.

Sometimes it's a matter of approach, some approach could be 5% better and 10x more expensive. So they will find the sweetspot, takes a few iterations.

Re: Grok3 Launch [video]

#522

Post by brulard »

liampulles wrote: post #187
phtrivier wrote: post #169

Off topic, but just in case: is there a good reference on how people actually use LLMs on a daily basis ? All my attempts so far have been pretty underwhelming: * when I use chatbots as search engines, I'm very quickly disappointed by obvious hallucinations * I ended up disabling github copilot because it was just "auto-complete on steroids" at best, and "auto-complete on mushrooms" at worst * I rarely have use cases…

I don't use it daily, and I find copilot counterproductive (for me). I do try to experiment with chatgpt when I remember to. I find it good for complex SQL, reviewing emails, and Godot assistance (I'm a beginner game Dev). There are also times when I have programming questions and I might try to use chatgpt, with mixed results. Our company has tried to integrate it into one of our products, and I find it troubling ho…

In the last few days I discovered it's good at medium-complexity SQL, not at really more complex ones. I'm struggling like 4th day with Claude, ChatGPT, Gemini and Deepseek. All could do some good analysis with some low-hanging fruit improvements, but all went completely crazy when trying to optimize more complex things, getting into loops proposing the same changes over and over, outputting invalid SQL, and gemini even forgot what we were doing, asking me if I could paste again the query i included in the very first message. Maybe the chain of thought models would handle this better, but I believe I hit the limit for the standard ones.

Re: Grok3 Launch [video]

#523

Post by weavejester »

sebzim4500 wrote: post #353
Mekoloto wrote: post #348

Earlier quoted context omitted.

Look who builds it. Its not Musk. Its his money who bought smart people. If he does to the USA Gov what he did to Twitter, he will destroy the brand, reduce the workforce by 80% and reduce the value by 80% too. The issue with him is, tha tin Twitter, the affected people had money. A missed payment of USA can literaly kill people.

Do you believe that Elon regrets acquiring Twitter? Despite being constantly told how much we was fucking up, it seems to have worked out OK for him.

It's lost 80% of it's value in 2 years, which usually isn't great. The most charitable view of X/Twitter is that it's now a propaganda platform that Musk doesn't mind taking a loss on in order to enact political change.

Re: Grok3 Launch [video]

#525

Post by gdhkgdhkvff »

Cheer2171 wrote: post #492
ban-evader wrote: post #436

Earlier quoted context omitted.

You’re a very poor user of LLMs if they’re not a net time saver for you.

So the No True Scotsman fallacy of LLM productivity?

Most people do see productivity gains from using LLMs correctly. Myself included. Just because some people don’t learn how to use them correctly doesn’t mean LLMs aren’t helpful. It’s like when internet search came out and a handful of laggards tried it once, failed to get the exact perfect result, and declared “internet search is useless”. Using a tool wrong is not evidence of the tool being useless, it’s evidence that you need to learn how to use the tool.

Re: Grok3 Launch [video]

#526

Post by spacebanana7 »

conradfr wrote: post #75
xqcgrek2 wrote: post #3

Looks impressive. OpenAI and Sam Altman might be cooked if its as capable as advertised.

Every competitors were done when Claude 3.5 was released, every competitors were done when o1 was released, the entire West was done when DeepSeek was released, the world was done when Mistral Le Chat was released, I guess now it's time for the solar system to be done because of Grok3. Let's see what new model dominates the galaxy next week.

There's a level of truth to many of those statements.

1) Claude 3.5 prevented OpenAI from making big monopoly profits on LLM inference 2) Open source models like Mistral and Llama effectively prevented any regulator from controlling how people fine tuned models, and what they used them for 3) Deepseek prevented the collective west from exerting control over the creation of base models

Re: Grok3 Launch [video]

#527

Post by trash_cat »

bhl wrote: post #250
phtrivier wrote: post #169

Off topic, but just in case: is there a good reference on how people actually use LLMs on a daily basis ? All my attempts so far have been pretty underwhelming: * when I use chatbots as search engines, I'm very quickly disappointed by obvious hallucinations * I ended up disabling github copilot because it was just "auto-complete on steroids" at best, and "auto-complete on mushrooms" at worst * I rarely have use cases…

I use it everyday. For coding, I use cursor composer to gather context about the existing codebase (context.md). Then I paste that into DeepSeek R1 to iterate on requirements and draft a high level design document, maybe some implementation details (design.md). Paste that back into composer, and iterate; then write tests. When I'm almost done, I ask composer to generate me a document on the changes it made and I doub…

Correct me if I am wrong, but the whole premise is of Cursor and Windsurf is that this architect-editor mode already being built in into the editor. This is why there is a distinction between composer (Editor) and chat function (arhitecture).

Re: Grok3 Launch [video]

#528

Post by jkestner »

guax wrote: post #510
dtquad wrote: post #478

Earlier quoted context omitted.

The full-sized DeepSeek-R1 is on par with o1. o1-pro is "o1 on steroids" and was the first selling point of the $200/month Pro subscription but they later also added "Deep Research" and Operator to the Pro subscription.

Every year seems like we get worse at naming things in non confusing ways. I am waiting for the o1-pro-max now, pro max ultra and pro max ultra plus.

Then you’ll know what the naming AI was trained on.

Re: Grok3 Launch [video]

#530

Post by _heimdall »

angusturner wrote: post #520

Credit to the engineers that built this, but it fills me with rage that Elon has this sort of unchecked power. How long before this starts getting deployed in safety critical applications or government decision making processes? With no oversight because Elon seems to have the power to dismiss the people responsible for investigating him. Anyone not scared by this concentration of power needs to pick up a book.

I'm not sure how LLMs/AI couldn't consolidate power. By design, it will move power from the individual to those running the AI systems.
Post reply on HN