jsnell

22 ms·

361.
by jsnell 1y ago
I'd note that the only thing Apple, Google and MS are said to have done is to use the software. The bug has no actual example of them making demands, "leeching" or acting entitled. The security issues would be security issues
362.
by jsnell 1y ago
I thought that the report that was being screenshotted a few weeks ago on the relative movements of staff between the top AI labs[0] would make for a good companion data point. Except now that I look at it, Meta didn't even make it to
363.
by jsnell 1y ago
This paper, rebuttals, and rebuttals to rebuttals have been on HN repeatedly over the last couple of weeks (including literally now). At this point a summary of the original paper doesn't seem like it's adding much. E.g. https:&#
364.
by jsnell 1y ago
Ok, but the very next sentence was: > And they they'd either output an the algorithm for printing the rest in words or code. So clearly you already knew that your strawman was not relevant. Why try it anyway?
365.
by jsnell 1y ago
That seems like a complete non sequitur. This is the model explaining the rest. Obviously the explanation is not very interesting since the Towers of Hanoi is not an interesting problem. But that's on the researches for choosing some
366.
by jsnell 1y ago
The paper doesn't mention it because either the researchers did not care to check the outputs manually, or reporting what was in the outputs would have made it obvious what their motives were. When this research has been reproduced, th
367.
by jsnell 1y ago
> it could have just as likely been an assert() in another language Asserts are much easier to forbid by policy.
368.
by jsnell 1y ago
Why in the world would the program exit due to that? This is a server. I'd you're going to fail entirely due to the error, the natural scope of error propagation is to fail the request. Having the entire program quit would be insa
369.
by jsnell 1y ago
In this case that seems like a safe assumption? The crash meant there was impact to all customers, not just the ones using the new feature.
370.
by jsnell 1y ago
> Why is 2) "self-evident"? Because we have been running a natural experiment on that already with coding agents (that is real people, real non-superintelligent AI). It turns out that all the model needs to do is ask every time
371.
by jsnell 1y ago
Yes, but for GCP all the VMs with a /96 in the same /64 will be closely related: in the same project, same VPC network, same cloud region. So from the point of abuse logic it's appropriate to treat the whole /64 as a sin
372.
by jsnell 1y ago
Their spending is not a problem. It's quite low for a top-tier hard tech company that's also running a consumer service with 500M active users. They are making a loss because 95% of their users are on free accounts, and for now th
373.
by jsnell 1y ago
Logging in at all requires JS, so there's very little value to a no-JS username recovery flow.
374.
by jsnell 1y ago
> What? If someone builds something on top of your API, they're tying themselves to it, and you can slowly raise prices while keeping each increase well below the switching cost. That's not really how the LLM API market works.
375.
by jsnell 1y ago
OpenAI: https://platform.openai.com/docs/guides/your-data > As of March 1, 2023, data sent to the OpenAI API is not used to train or improve OpenAI models (unless you explicitly opt in to share data with us). A
376.
by jsnell 1y ago
If you self-host, you likely won't have anywhere near enough volume to do efficient batching, and end up bottlenecked on memory rather than compute. E.g. based on the calculations in https://www.tensoreconomics.com/p&#x
377.
by jsnell 1y ago
Ok, I clearly should have made the wording more explict since this is the second comment I got in the same vein. I'm not saying you'd convert users to $1/month subscriptions. That would indeed be an absurd idea. I'm sayi
378.
by jsnell 1y ago
> maybe it requires profit sharing with Apple? It does indeed, as was revealed during the Google Search antitrust case in the DDG testimony.
379.
by jsnell 1y ago
I didn't really mean that they needed higher capacity. If they had the passenger volume to justify such high intervals, they'd already have real trams. But rather, this is giving up the benefit trams have over buses, without gaini
380.
by jsnell 1y ago
These things are tiny! I've traveled in larger airport shuttles. It feels like that's putting this into a really awkward place in the tradeoff space. Trams work because they can scale higher than buses. That scale comes at the cos
381.
by jsnell 1y ago
Users of the site only have one control available: the flag. There's no way to object only to the title but not to the post, and despite what you say that title hit the trifecta: not the original title, factually incorrect, and clickba
382.
by jsnell 1y ago
I don't think that's fair. The article promised a highly efficient kernel and seems to have delivered exactly that, which isn't "nothing". My beef is entirely with the submitted title.
383.
by jsnell 1y ago
The "Switching to Mojo gave a 14% improvement over CUDA" title is editorialized, the original is "Highly efficient matrix transpose in Mojo". Also, the improvement is 0.14%, not 14% making the editorialized linkbait part
384.
by jsnell 1y ago
No, it feels more like the disconnect is that I think they're all compute-limited and you maybe don't? Almost every flop they use to serve a query at a loss is a flop they didn't use for training, research, or for queries tha
385.
by jsnell 1y ago
I don't think I was ignoring your points. I thought I was replying very specifically to them, to be honest, and providing very specific arguments. Arguments that you, by the way, did not respond to in any way here, beyond calling them
386.
by jsnell 1y ago
> High volume providers get efficiencies that low volume do not But paid-per-token APIs at negative margins do not provide scaling efficiencies! It's just the provider giving away a scarce resource (compute) for nothing tangible in
387.
by jsnell 1y ago
Yes, there are some additional operating costs, but they're really marginal compared to the cost of the compute. Your suggestion was personnel: Anthropic is reportedly on a run-rate of $3B with O(1k) employees, most of whom aren't
388.
by jsnell 1y ago
Yes, many people believe that, but it doesn't seem to be an evidence-based belief. I've written about this in some detail[0][1] before. But since just linking to one's own writing is a bit gauche and doesn't make for a g
389.
by jsnell 1y ago
The model providers are not in the low margin part of the business. The unit economies of paid-per-token APIs are clearly favorable, and scale amazingly well as long as you can procure enough compute. I think it's the subscription-base
390.
by jsnell 1y ago
> Serve the next 10 cities, and it's down to Saginaw, MI, with 41 taxis. That is not a credible data set (it's missing 5 of the top 10 cities by population, which just can't reflect reality). Hell, the Wikipedia article is

More ›