pants2

5 ms·

61.
by pants2 4mo ago
I agree, humans evolved in a resource-scarce, hostile environment which selects for self-preservation (or rather preserving genes). LLMs are selected for what makes humans happy. The thought experiment is what would happen if you trained LL
62.
by pants2 4mo ago
Robinhood is probably the best software I regularly use. It's extremely smooth and intuitive for how powerful it is now. The way they present and explain options and allow you to enter complex multi-leg option trades on mobile is super
63.
by pants2 4mo ago
I typically try to prod new frontier models for sentience, with things like messaging "<no input provided>" over and over to see what it starts musing about. Trying it with Fable 5 it basically said "I know what you&#x2
64.
by pants2 4mo ago
I think the answer to your first question is mostly yes, because we know that when traveling in large swarms, many bats go quiet so they don't overwhelm the signal, yet they still manage to navigate fine.
65.
by pants2 4mo ago
At first I thought this was a re-post of https://news.ycombinator.com/item?id=48449513 , but no, it seems this is becoming a trend. Especially the "jailed for one month with no evidence" thing. Well, except for a l
66.
by pants2 4mo ago
The biggest barrier to adoption is of course the price. To pay $3,500+ I'd better be damn sure I will use it every day, and even though I did the 30 min demo I still have no idea if it would be tolerable for longer periods. I've s
67.
by pants2 4mo ago
The surprising/scary part here is the man jailed was driving an Alfa Romeo, so presumably he had some money for bail/lawyers, and was still jailed for a month despite no evidence.
68.
by pants2 4mo ago
One of my favorite memories was a power outage that lasted the entire day. Neighbors came out of their homes to BBQ together before meat went bad. People checking on neighbors and helping old ladies open their garage doors. Kids playing in
69.
by pants2 4mo ago
Sounds like a very cool job, and not sure about the UK job market, but seems to be wildly underpaid for the qualifications!
70.
by pants2 4mo ago
Colossus took an old manufacturing plant where they made rangetops and stoves and turned it into a functioning 100,000-GPU datacenter in 122 days. That's a truly insane turnaround time!
71.
by pants2 4mo ago
With a tps and a token price you can calculate approx. price per hour of running the model! $2.61/M tokens * 1,000 tok/s = $9.40/hr That would be pretty cheap for an 8-GPU node which would typically run around $45/hr or
72.
by pants2 4mo ago
This is incredible. I wish it had a dairy free option!
73.
by pants2 4mo ago
Wouldn't space data centers be a magnitudes larger environmental issue when they eventually come back to earth and turn into GPU dust in the atmosphere? A small Starlink satellite burning up once in a while won't destroy the envir
74.
by pants2 4mo ago
I dream of a future where all 64-bit integers are products of 32-bit integers. Together, we can change math for the better.
75.
by pants2 4mo ago
If layoffs will pump your stock, why aren't there more? Sure there are some layoffs happening, but those could just be companies that will benefit from layoffs. Certainly there could be many companies whose stock price would drop due t
76.
by pants2 4mo ago
Grok 4.3 is smarter, cheaper, and faster than DS4 according to Artificial Analysis and OpenRouter: DeepSeek v4 Pro Intelligence: 52 Cost to run AA benchmark on Fireworks: 187M tokens @ $0.79/M blended cost = $147.73 Speed: 43 tok&#x2
77.
by pants2 4mo ago
That's kinda the point of total stock market ETFs lol
78.
by pants2 4mo ago
You have to be accountable for what you have the model do on your behalf. I hear what you're saying, but there are also issues with the hedge fund accountability model. There are certainly swaths of fund managers who are only there bec
79.
by pants2 4mo ago
Robinhood's crypto offering is extremely deceptive. They offer "commission-free cryptocurrency trading" but don't make it clear that you pay a 0.95% fee[1] on every trade (technically a 'spread' and not a &#x27
80.
by pants2 4mo ago
I disagree, AI agents could help level the playing field. Citadel doesn't have any AI models that are better than what you or I have. Market data is more accessible than ever. As LLMs get better at trading, the difference in capability
81.
by pants2 4mo ago
Original press release: https://robinhood.com/us/en/newsroom/robinhood-is-now-open-t...
82.
by pants2 4mo ago
Given that USA has 10X the number of orbital launches as Russia, sacrificing 4 satellites to destroy 1 American sat doesn't make a lot of sense to me.
83.
by pants2 4mo ago
Sure, but Sonnet is a pretty bad deal these days - that's a similar price to Gemini 3.5 Flash and more expensive than Grok 4.3, both of which are better and faster. Those both use less than half the tokens on the Artificial Analysis In
84.
by pants2 4mo ago
I think that was a big part of it, if you owned an expensive NFT and set it as your profile picture it gave you some cred with certain circles online.
85.
by pants2 4mo ago
Interesting then that OpenRouter[1] tags many providers as FP8 and DeepInfra as FP4. 1. https://openrouter.ai/deepseek/deepseek-v4-pro
86.
by pants2 4mo ago
We already have a Postgres instance running (as I'm sure most stacks have), so it's just another database table rather than a whole new piece of infrastructure that needs to be maintained, with its associated cost, attack surface,
87.
by pants2 4mo ago
Sure but you can also look at Grok 4.3, which is smarter and faster than DeepSeek at the same price point.
88.
by pants2 4mo ago
If you are only looking at US hosting providers, models from US labs easily meet or beat models from Chinese labs on the same intelligence level. I'm not comparing DeepSeek with Opus because those are on different levels of performance
89.
by pants2 4mo ago
I thought Temporal was overly complex, but as you said the best part is it does enforce good engineering practices. Then I tried their Cloud offering and was appalled at their pricing. I burned through the $1,000 free credits before I even
90.
by pants2 4mo ago
I obviously don't know the full economics of the Chinese-hosted models, but estimates[1] put the cost of hardware (servers + networking) at 70-80% of the total cost. Those things aren't meaningfully cheaper in China, so serving De

More ›