redox99

13 ms·

241.
by redox99 4mo ago
Because having rules on HN comments means users adapt their writing to those rules. Whereas the same on external content means potentially missing out on interesting stuff. We've already seen big announcements being clearly AI generate
242.
SpaceX has just officially unveiled its AI1 satellite(twitter.com)
6 points by redox99 4mo ago|0 comments
243.
by redox99 4mo ago
I don't see Alphabet share price changing much just because of SpaceX being valued 2T instead of lets say 1T (being extremely generous). In fact this deal will hurt their profits, which is more likely to hurt Alphabet stock price than
244.
by redox99 4mo ago
You seem to imply that with this deal their shares are worth 88B but without it they're worthless. It's very hard to know how much the deal actually increases SpaceX market cap, but unless Google exits their SpaceX position soon i
245.
by redox99 4mo ago
That's a good point. That makes it less likely that it's training, but who knows.
246.
by redox99 4mo ago
That's for standalone servers. xAI has something that nobody else has. The single largest interconnected cluster[1]. For inference it doesn't really matter, but for training that networking is crucial. [1] Probably, there could be
247.
by redox99 4mo ago
People definitely wanted to train deep networks before, but didn't know how. They evdn tried things like training layers independently. I don't think it was intuitive to anyone back then, the vanishing gradient problem was a big d
248.
by redox99 4mo ago
Almost everything in ML is like that. It seems so obvious in hindsight. It's maybe what I love most. Residual connections are so simple, so obvious and so vital. Yet nobody came up with them until 2015?
249.
by redox99 4mo ago
In the 90s you didn't have norm layers, residuals, attention, and some more. So you're missing a lot of the building blocks that make LLMs. It's not a matter of just having the compute.
250.
by redox99 4mo ago
Chain of Thought was kind of an obvious solution that everybody knew was necessary by the time chatgpt / gpt4 came out. It was just a matter of time that frontier labs actually shipped it. MoE was also pretty straightforward, just a bi
251.
by redox99 4mo ago
It can answer presidents fine. It fails for vice presidents. ----------- As of my last update, here are the five most recent individuals to have served as Vice President of Argentina: Sergio Massa (Served as Vice President from 2019 to 2023
252.
by redox99 4mo ago
Argentina vice presidents span from 2007 to 2023. Knowledge cutoff cant explain getting all 5 of them wrong.
253.
by redox99 4mo ago
I tried most prompts that didn't rely on recent knowledge on the basic "AI Chat", not the "Agent skills" version. I just tested "List the 5 most recent Argentina vice presidents" on E4B and it literally go
254.
by redox99 4mo ago
Yes, but as Elon himself remarked, these are short term leases (90 day exit window), not guaranteed long term contracts.
255.
by redox99 4mo ago
I agree that the web search tool probably is pretty bad. However a smart model would never hallucinate impossible weather data if the search tool failed. I'm sure you can get some out of it if you babysit it with an optimized prompt, h
256.
by redox99 4mo ago
I think Gemma 4 12B is definitely possible to run on high end phones, google claims you need 16GB of memory. But it's probably not very usable, you'll need to swap most stuff other than the LLM. When I tried E2B and E4B with Googl
257.
by redox99 4mo ago
I was just testing Gemma E2B and E4B yesterday, and they are just too dumb to be useful outside of niche use cases. Besides, there's no good agent on Android. Having a model that can't run web searches and browse websites is limit
258.
by redox99 4mo ago
Energy cost can be theoretically lower than on earth IF launch costs and additional hardware costs are low enough. Because solar is about 5x better on space than on earth. That's basically it.
259.
by redox99 4mo ago
My back of the napkin math says you need launch prices of about $100/kg to $200/kg for it to make sense. The cooling problem is vastly exaggerated, you need around 0.5x the area of your solar panels in radiators. I think AI infere
260.
by redox99 5mo ago
I think it's cool how in a decade we went from "Neural networks will never be able to understand this sentence that's obvious to humans" to "LLMs must be able to solve problems that humanity hasn't been able to
261.
by redox99 5mo ago
> Think of it as 1000+ ways to think of that word/token I assume you used 1000 because that's in the ballpark of the vector size. But these are not independent scalars, like each might store a certain property. Just like in 2D
262.
by redox99 5mo ago
Do questions like goldbach conjecture, fermats last theorem, etc deserve more credit than whoever answers them?
263.
by redox99 5mo ago
Instead of going into a long technical argument of why your description of LLMs is flawed, I'll go straight to the point, because people keep moving the goal posts. What exact problem would need to be solved by LLMs to convince you tha
264.
by redox99 5mo ago
Data centers don't grow on trees.
265.
by redox99 5mo ago
You're partly right, but we can't ignore that EUV is inherently a lot more expensive, and progress towards new nodes is very slow (current fast hardware will stay fast much longer). Back in the day the PS3 came with a 90nm GPU and
266.
by redox99 5mo ago
Now that Moores law is dead I think GPUs will depreciate a lot slower. I mean there's already a lot of hardware that has gotten more expensive in the last 5 years.
267.
by redox99 5mo ago
Interesting, that definitely seems to be it.
268.
by redox99 5mo ago
You were getting 1k rps, and each request was from an unique IP? So after an hour you got hit by 3.6M different IPs? And all from uncorrelated /16s? That seems hard to believe. Not that I don't believe you, it's just hard for
269.
by redox99 5mo ago
That's 16.6rps. A single guy holding the F5 key on chrome can generate that much traffic and take down your website. That kind of performance was never acceptable.
270.
by redox99 5mo ago
Probably you don't have much to scrape?

More ›