jsnell

31 ms·

421.
by jsnell 1y ago
The suggested attack was the attacker writing an infinite stream of EHLOs on a connection. What's the scenario where an attacker has full control of the SMTP control framing, but doesn't have attack to the payloads?
422.
by jsnell 1y ago
What's the threat model here? A scenario where the attacker controls the plaintext being sent but doesn't know what the plaintext is seems quite unlikely.
423.
by jsnell 1y ago
There are multiple search engines known to be based on Google's API (Startpage, Leta, Kagi), so that product definitely exists. But it exciting that's all we know. They indeed do not publish anything about it. We don't know t
424.
by jsnell 1y ago
Yes. The Bing Search API is priced at $15/1k queries in the cheapest tier, Brave API is $9 at the non-toy tier, Google's pricing for a general search API is unknown but their Search grounding in Gemini costs $35/1k queries. S
425.
by jsnell 1y ago
> Through formalized infrastructure collaborations, and in coordination with the US government, OpenAI will: > Partner countries also would invest in expanding the global Stargate Project—and thus in continued US-led AI leadership and
426.
by jsnell 1y ago
The text in the model card says the results are from March (including the Gemini 2.5 Pro results), and o3 wasn't released yet. Is this maybe not the updated card, even though the blog post claims there is one? Sure, the timestamp is in
427.
by jsnell 1y ago
There's nothing intrinsically intuitive about letter grades. It's just that you've been taught those specific arbitrary mappings. Consider for example the "S" as a better grade than "A", originating from J
428.
by jsnell 1y ago
For me the really unreasonable change was the app testing requirements on non-corporate developers. Having to get 20 users to beta test an unlisted Android app for two weeks before getting it on the store is not a reasonable thing to requ
429.
by jsnell 1y ago
You can enter a prompt from the "customize" dialog. Have you tried asking for a more specifics, assume the audience is an expert on the subject, and cut down on the fluff?
430.
by jsnell 1y ago
> To steelman it, the US has a far lower dependence on trade as a percentage of GDP than just about any other country, so if any country will be likely to (raw-economic-numbers) survive a shock like this, it's the US. The reason thi
431.
by jsnell 1y ago
But the hypothetical part here isn't just the use case. It's also the outcome. And the bad outcome did not happen, the author made it up.
432.
by jsnell 1y ago
> Uber doesn't really have any way to grow. That's demonstrably not true. Their latest earnings (Q4 2024) showed a 20% growth in revenue, an 18% increase in trips, and a 14% increase in the number of monthly active users (all n
433.
by jsnell 1y ago
> Did Zuck buy Insta to neutralize a competitor? Sure seems like it! For one thing, Zuck cancelled all work on Facebook Camera "since we're acquiring Instagram." I don't get this bit. That seems to be supporting exact
434.
by jsnell 1y ago
20 years ago I think the coverage area of Google Maps was still strictly limited to the US + UK. Like, the rest of the world map was empty. But I worked on Google's geocoding (mapping names to locations) and reverse geocoding (mapping
435.
by jsnell 1y ago
> They could have cited the industry standard to block third party cookies in other browsers and done so without a replacement, the reason they are being prohibited from doing so is because they are motivated to maintain data access for
436.
by jsnell 1y ago
It being a language feature gives you controlled access to the lexical scope, such that the template string can refer to variables by name rather than having to pass each value as a parameter. Doing it via parameters is repetitive and/
437.
by jsnell 1y ago
This number does not seem credible. Most likely Altman just made it up. Back of the envelope: OpenAI inference costs last year were 4B. Tens of millions would be at least 20M, i.e. 0.5%. That 4B is not just the electricity cost. It needs to
438.
by jsnell 1y ago
Memory is needed for more than just the parameters, e.g. the KV cache.
439.
by jsnell 1y ago
No, I was pretty careful to address only the very specific claim the OP made about how effective AI use is not a skill. If you're reading anything more than that into the comment, I think you're projecting. I really don't car
440.
by jsnell 1y ago
Of course using AIs is a skill, just like e.g. effectively writing search queries used to be a skill back in the day. When I actually tried getting something done with AI models for the first time, rather than just kicking the tires with th
441.
by jsnell 1y ago
It does, that point in the tradeoff space was not available until now. Any model that's not dominated by at least one model on both axes will push forward the frontier. (The actual frontier isn't actually a straight line between t
442.
by jsnell 1y ago
I clearly didn't understand what you meant, because you did in fact have additional unstated requirements that I could not even have imagined existed and were not in any way hinted at by your initial spec. And I still don't know w
443.
by jsnell 1y ago
No. If you're able to verify correctness of individual items of work, you can accept the 93% of verified items as-is and send the remaining 7% to some more expensive slow path. That's very different from just knowing the aggregate
444.
by jsnell 1y ago
Even as a human programmer I don't actually understand your description of the problem well enough to be confident I could correctly guess your intent. What do you mean by "highlight as you scroll"? I guess you want a single
445.
by jsnell 1y ago
> I wish I could try the AI separate from my Google account. If that's a concern, just create another account. Doesn't even require using a separate browser profile, you can be logged into multiple accounts at once and use the
446.
by jsnell 1y ago
The free-with-small-rate-limits designator was "experimental", not "preview". I think the distinction between preview and full release is that the preview models have no guarantees on how long they'll be available
447.
by jsnell 1y ago
> This implies it's not a hybrid model that can just skip reasoning steps if requested. It clearly is, since most of the post is dedicated to the tunability (both manual and automatic) of the reasoning budget. I don't know what
448.
by jsnell 1y ago
Apple has fewer employees than Google (164k < 183k).
449.
by jsnell 1y ago
I think they should name them after fictional characters. Bonus points if they're trademarked characters. "You gotta try Mickey, it beats the crap out of Gandalf in coding."
450.
by jsnell 1y ago
Honestly, this reads like an AI-generated summary. Discussion on original paper: https://news.ycombinator.com/item?id=43086347

More ›