jsnell

15 ms·

211.
by jsnell 1y ago
Sure, let's compare. Apple's product revenue in this fiscal year has been $233B, with a gross margin of $86B. Their services revenue is $80B with $60B gross margin.
212.
by jsnell 1y ago
In this case there is no risk of anyone stealing Intel's ideas or even reacting to them. First, they're not even an also-ran in the AI compute space. Nobody is looking to them for roadmap ideas. Intel does not have any credibility
213.
by jsnell 1y ago
There are multiple ways to do the computation. All of them will show LLMs having unit economics that are at least an order of magnitude better than search engines for the search engine use case[0]. Not multiple orders of magnitude worse
214.
by jsnell 1y ago
6 months seems like a carefully cherry-picked number there, it is basically middle of the the trough of the tariff-induced market collapse that started 7 months ago and had been recovered from 5 months ago. That explains about 20 percentage
215.
by jsnell 1y ago
Their terms are that you can't sell *Steam keys* for cheaper than the game is listed for on Steam. There is a class-action lawsuit on this that's been ongoing for half a decade now, but as far as I can tell the plaintiffs have not
216.
by jsnell 1y ago
> None of this is to say there’s a problem with this - it’s just fascinating that net result of this huge spike is lots of jewelry and old coins being melted into bricks of gold that will sit quietly in secure warehouses and vaults aroun
217.
by jsnell 1y ago
It doesn't need to be new ad spend. Just the possibility of the existing ad spend being up for grabs justifies all the capex so far.
218.
by jsnell 1y ago
> Where I suspect LLM output won’t help much. [...] Low level infrastructure code the kind I’ve spent my whole life on That's just wishful thinking. First, that kind of code is one of the things that's most valuable for AI to g
219.
by jsnell 1y ago
That's 1/6th of Anthropic, not half.
220.
by jsnell 1y ago
No relation AFAIK, this is Mira Murati's post-OpenAI joint. Which makes the hero image on this page being a diagram of Danny Hillis' TinkerToy computer all the more baffling.
221.
by jsnell 1y ago
> Getting the most out of models requires maximally exploiting parallelism, so sandboxing is a must. What are your thoughts on checkpointing as a refinement of sandboxing? For tight human/llm loops I find automatic checkpoints (of b
222.
by jsnell 1y ago
Doing this as a social app somehow feels really gross, and I can't quite put to words why. Like, it should be preferable to keep all the slop in the same trough. But it's like they can't come up with even one legitimate use
223.
by jsnell 1y ago
Probably caught by the flamewar detector (too many comments compared to upvotes).
224.
by jsnell 1y ago
But what the OP wrote was not "check regularly and taste". They proposed a single timer-based check at one minute less than the time on the box. That strategy relies on the box being off by at most one minute, so the results from
225.
by jsnell 1y ago
I thought the original article included the strongest objective data point on this: recent progress on the METR long task benchmark isn't just on the historical "task length doubling every 7 months" best fit, but is trending
226.
by jsnell 1y ago
> but then why are we comparing such distinct and unrelated tasks as ... Because a few years ago the LLMs could only do trivial tasks that a child could do, and now they're able to do complex research and software development tasks.
227.
by jsnell 1y ago
It's the only benchmark I know of with a well-behaved scale. Benchmarks with for example a score from 0-100% get saturated quite quickly, and further improvements on the metric are literally impossible. And even excluding saturation, t
228.
by jsnell 1y ago
Indeed you can't be sure. But on the other hand a bunch of the commentariat has been claiming (with no evidence) that we're at the midpoint of the sigmoid for the last three years. They were wrong. And then you had the AI frontier
229.
by jsnell 1y ago
The key problem is that his economic analysis is absolute trash. I used to think he was just totally incompetent at it, but given the bias in the errors, it is pretty clearly intentional deception. But it's often pretty hard to address
230.
by jsnell 1y ago
Interesting if he really is more convincing in speech than in writing. His writing is certainly flamboyant, but the aggression and expletives seem more targeted at hyping up people who already believe the things he writes, not for making pe
231.
by jsnell 1y ago
1.21GW is an absurd level of precision for this kind of prediction.
232.
by jsnell 1y ago
Their process is documented at https://opensource.google/documentation/reference/releasing So it could even be a pure hobby project - not something done for work - where the initial author (over a decade ago) chos
233.
by jsnell 1y ago
But then Facebook would have had no incentive to offer the agreement to her. And she must have been very happy with what she got in the deal, given she gave up the option of talking about all the apparently horrible misdeeds she witnessed w
234.
by jsnell 1y ago
Google's non-advertising revenue is about $25B/quarter. There's only a handful of tech companies with higher revenue than that. It is for example a bit more than Tesla, double that of Broadcom or Oracle, a bit less than Dell
235.
by jsnell 1y ago
What you're ignoring is that this was a change to an ad blocker[0], not a change to the site . Google did not implement a change to stop counting views. An ad blocker intentionally[1] choosing to block the long-standing API calls used
236.
by jsnell 1y ago
The comment you replied to explained that nothing was changed on the YouTube side. This was an adblocker choosing to start block a non-ads, non-tracking, samesite API call that had probably been in place for like a decade. So it's quit
237.
by jsnell 1y ago
A terse prompt and expecting a one-shot answer is really not how you'd get an LLM to solve complex problems. I don't know what Deepmind and OpenAI did in this case, but to get an idea of the kind of scaffolding and prompting strat
238.
by jsnell 1y ago
This is how it has worked for ages. If you think about it for a bit, I think you'd come up with all kinds of reasons for why this can't be done with just server-side signals. For example, how do you account for skipping over alrea
239.
by jsnell 1y ago
The headline seems pretty aspirational. The licensing standard they're talking about will achieve nothing. Anti-bot companies selling scraping protections will run out of runway: there's a limited set of signals, and none of them
240.
by jsnell 1y ago
https://news.ycombinator.com/item?id=30757889

More ›