jsnell

19 ms·

271.
by jsnell 1y ago
No, the economics will never work out for a Proof of Work-based counter-abuse challenge. CPU is just too cheap in comparison to the cost of human latency. An hour of a server CPU costs $0.01. How much is an hour of your time worth? That&#x2
272.
by jsnell 1y ago
Probably Google Zanzibar (and the various non-Google systems that were created as a result of the paper describing Zanzibar).
273.
by jsnell 1y ago
> Hard to count the number of things that can go wrong by relying directly on an LLM to manage audit/activity/etc. logs. Nothing in this post suggests that they're relying on the LLM itself to append to the audit logs. Tha
274.
by jsnell 1y ago
The actual report (which this article doesn't link to; bad Axios): https://nanda.media.mit.edu/ai_report_2025.pdf
275.
by jsnell 1y ago
The 9% number comes from dividing by the number of IPv4 hosts reported by Censys, who do a portscan of the entire IPv4 space. But obviously most clients will not have any ports open, and wouldn't be visible to the scan. It's not a
276.
by jsnell 1y ago
> Content-Length+Transfer-Encoding should be bad request. Maybe it should be, but it isn't. In fact, all revisions of the HTTP/1.1 RFC have made it clear that if both headers are present, the receiver must treat it as the Conte
277.
by jsnell 1y ago
The author is only arguing against HTTP/1.1 for use between proxies and backends. Explicitly so: > Note that disabling HTTP/1 between the browser and the front-end is not required
278.
by jsnell 1y ago
The duration you'd need the memory for is a couple of seconds, during which time you're pegging a CPU core on the computation anyway. It is not needed for the entirety of the browsing session. Now, could you construct a challenge
279.
by jsnell 1y ago
No, but what I'm saying is that these scrapers are already not using GPUs or ASICs. It just doesn't make any economical sense to do that in the first place. They are running the same Javascript code on the same commodity CPUs an
280.
by jsnell 1y ago
A different algorithm would not help. Here's the basic problem: the fully loaded cost of a server CPU core is ~1 cent/hour. The most latency you can afford to inflict on real users is a couple of seconds. That means the cost of pa
281.
by jsnell 1y ago
This was beyond predictable. The monetary cost of proof of work is several orders of magnitude too small to deter scraping (let alone higher yield abuse), and passing the challenges requires no technical finesse basically by construction.
282.
by jsnell 1y ago
This is an LLM-generated summary of other documentation, right?
283.
by jsnell 1y ago
For the first couple of pages I thought that the article was a gag, and {C,S}-Insert was going to be the punchline.
284.
by jsnell 1y ago
The unit economics of inference are incredibly good. To be profitable, OpenAI would need to be 1/10th as good as Meta or Google are at monetizing via ads.
285.
by jsnell 1y ago
The reason I'm asking this is exactly their blog post, which says says that their goal is to roll this out to 50% of the French queries at the end of the year. That's a rather limited slice of traffic (50% of French queries, 0% of
286.
by jsnell 1y ago
How do you know those good-looking results were served from the new index, rather than being just rebranded Bing API results?
287.
by jsnell 1y ago
> These days, LLMs are what we call "Mixture of Experts", meaning they only activate a small subset of their weights at a time. This makes them a lot more efficient to run at high batch size. I don't really understand why
288.
HTTP/1.1 must die: the desync endgame(portswigger.net)
3 points by jsnell 1y ago|0 comments
289.
by jsnell 1y ago
Does sufficient capacity for recasting actually exist in the US or would it need to be built? If the latter, how long will it take to build it? I suspect it doesn't exist yet. In fact, the majority of the US trade deficit with Switzerl
290.
by jsnell 1y ago
Yeah, the interaction model around pushing commits is absolutely baffling.
291.
by jsnell 1y ago
Nice! You definitely have something there. I would love a dictionary lookup. I'd expect a big part of the strategy to be working backwards from the possible 1/2 letter words, but there's no way to tell how reasonable the di
292.
by jsnell 1y ago
> 10x productivity means ten times the outcomes, not ten times the lines of code. This means what you used to ship in a quarter you now ship in a week and a half. Not really? That's defining productivity as latency, but it's at
293.
by jsnell 1y ago
> now look at the actual pricing history of frontier models, the ones that 99% of the demand is for at any given time: The meaningful frontier isn't scalar on just the capability, it's on capability for a given cost. The highes
294.
by jsnell 1y ago
Fair enough, sorry about that. You called it "tampering" and "alteration". But honestly I don't see any distinction to "editing". All of those imply that the email is being changed. It isn't. The emai
295.
by jsnell 1y ago
Why in the world are you and the op calling this "editing"? Like, the OP clearly does not understand the distinction between rendering the email and storing it, given their "evidence". But you do understand the differenc
296.
by jsnell 1y ago
XAI does not include Twitter, as far as I know? Aren't they separate entities? But Twitter can't be worth more than a couple of tens of billions at the most. Things OpenAI has that XAI doesn't: Hundreds of millions, if not a
297.
by jsnell 1y ago
Information retrieval for consumers is an existing high margin $250B/year business at a minimum (search ads revenue). That market is being disrupted before our eyes. That's 20x right there. There's going to be fierce competit
298.
by jsnell 1y ago
I find it strange how most of these terminal-based AI coding agents have ended up with these attempts at making text UIs flashy. Tons of whitespace, line art, widgets, ascii art and gradients, and now apparently animations. And then what yo
299.
by jsnell 1y ago
That is not true.
300.
by jsnell 1y ago
I don't think this quite works. It seems like the simple strategy is to carve off a fully contained 1x2 "room" on the bottom row for the red dot, play normal Tetris except make sure you never clear the 2nd lowest row, and the

More ›