And the most reliable. A year and a half ago this was the exact reason why I switched to Gemini API from OpenAI. OAI would make me wait up to 10 minutes for the first token in the most egregious cases. Gemini served me almost immediately every time.
Huh? The constant failures in the Gemini API and overnight changes are well documented, there was a time from January to roughly July (maybe it still persists) where the API constantly spew 503 errors. This was so bad that I switched to OpenAI API instead. See e.g. https://discuss.ai.google.dev/t/frequent-503-errors-service-...
Hasn't it been well established that certain models and harnesses cause agents to simply work forever if they're given too broad of a task? I think that that is what happened here. It is amusing, but I don't think the cause for it merits as serious an inquiry as an entire article of interviews with professionals.
MIT License requires 'attribution', e.g. at the very least the original copyright notice and license text (which tbf is the case in aloshdenny's copy because he left the LICENSE.md intact (or just forgot to "clean up").
this is completely and totally incorrect. MIT lets you redistribute, copy, and modify. It does not let you take authorship over the original work, even if just by way of changing git history.
In the US, claiming authorship isn't as protected or significant as internationally; moral rights are only protected for "works of visual art". https://en.wikipedia.org/wiki/Moral_rights
Which is not to say that uploading something to github is necessarily asserting ones authorship (if you read the front matter of a book, you might see a phrasing such as "[Name] asserts the moral right to be identified at the author of this work."; this is intended for jurisdictions that comply with the Berne convention more fully than the US does.) It is also not an instrument of conveyance (which needs to be signed by the owner of the rights) or the same as registering a copyright. A copyright notice is no longer required since the Berne Convention Implementation Act of 1988 (effective as of 1989), but would involve the copyright symbol and not a git upload.
This is jurisdiction specific. The US doesn't have anything like this as long as you're not claiming to own the copyright. As far as I know, this is also why in some European countries you can't simply make a work public domain, as you can't give up attribution rights. In the US, you just don't have any attribution rights (though you could demand attribution as part of a copyright license).
And in any case, the copyright notice is still intact. They did attribute the author.
This is untrue and if you thought about what you're claiming for a second this should be obvious. The MIT license let's someone do any of the things outlined there subject only to the need to retain the copyright notice and the license itself. I can grab MIT-licensed files and commit them to my repo (e.g., vendoring a dependency). I can also grab the files from an MIT-licensed commit and then the files from the next commit and recreate the history of that repo with myself as the author provided I've met those two conditions. I'm not saying this isn't _scummy_ behavior; it is! But its within the bounds of what the MIT license allows.
Waive. And no, it doesn't but it doesn't require attribution either. So you're falling back on copyright law which is unlikely to protect you here - there's a reason people include licenses, after all.
People include licenses because if there isn't any license copyright law applies and you may not create any copy of said code. You as a third party have no use of the code without a license
Software licenses aren't enforced by magic but they are social contracts: even if nobody has the resources to go after you for doing it, it is pretty clearly against the permissive licenses' spirit to misrepresent yourself as the author
It's also pretty clear that the real author gave you permission to misrepresent yourself as the author, as long as you retain the copyright notice. If he didn't want you to do that, he shouldn't have given you permission.
The "if they wanted you to follow the spirit, the letter should have been different" argument grants membership in the society of people known throughout history as "the reasons we can't have nice things"
Incorrect, there are plenty of people who ignore or fail to recognize the purpose of a rule as a completely free service, as any forums mod or OSS project maintainer can attest to
they dont though. nothing about MIT license is granting anyone the ability to misrepresent as the author. thats why its a license and not copyright transfer.
thats just not true. Commits themselves are held up as copyrighted and a claim for creation. just as you cant legally put up a poster claiming you painted a painting you just got a license for, you can't claim the commits.
What a snooze fest. Another model that does not meaningfully improve on intelligence or price compared to its peers. Google has basically announced that they've "caught up" with the rest. I think they've been doing great work in the Flash department so seeing this is... underwhelming?
The only models beating it are on "max", while this is "high". There's no guarantee that those effort / reasoning levels compare, but there's almost certainly an "xhigh" or "max" version later which will score higher there.
I found that Google does not benchmaxx as much as the other providers. Of you look at real-case evaluation like lm-arena, even the 3.8 flash is often near the top despite its benchmark index being worse.
I noticed this recently with the user submitted benchmarks on Kaggle. For these obscure tests that the models haven't seen, Gemini 3.8 is often on par or beating other frontier models. Gemini is a bit shite at the game checkers though, for some weird reason it performs poorly on those benchmarks.
To play the Devil's advocate, Claude loves chugging tokens, while Gemini appears to be quite a bit more conservative and efficient. I believe AA's price per task breakdown reflects this.
I have never seen a multi-billion $ company that has such... bizarre? conduct. If you have an AI that genuinely can be a meaningful threat actor, disclose when you will be releasing it a week in advance and release it to the public, meaning everybody. Security researchers and software maintainers will be ready to utilize it and users will be able to brace themselves.
This makes sense. I'm assuming that, for the AI search assistant, Google has worked plenty to refine data lookup and retrieval—they definitely have some internal tech that competitors don't.
To put this into perspective, healthcare, real estate and banking each earn ~$1.5T in revenue annually. This is REVENUE, not profit which is obviously lower than the revenue in every industry. I fail to understand how AI can achieve $6T in annual revenue when the TAM for healthcare and banking is a quarter of that in the US.
The article says $6T by 2031 and it's global spending, not just in the US.
> Bain forecasts that annual spending on AI infrastructure might hit $1.5 trillion by 2031.
So maybe in 5yrs it will need enough revenue to cover $1.5T in capex spend globally in a year, investments which as the article mentions is also being supported by governments in "the UAE, Saudi Arabia, the EU, South Korea and the US". ... Or spending growth could lower over the next couple years to be more realistic.
None of this refutes my point which was that there's no way that the AI industry can, even globally, achieve a $6T TAM. At the very least, I'm confident that it will not happen by 2031. Industries cannot grow that quickly because money can't reach them that quickly.
I'm fairly confident that the crash will happen because token prices fall through the floor due to competition from the open source models. This will destroy the valuation of the AI firms and their ability to raise capital, so they'll stop spending hundreds of billions per annum on training.
OTOH, this price crash will increase demand for inference. Maybe some data centres will go bankrupt, but the scavengers won't be selling off used TPU's for cheap, they'll operate those data centers to sell you tokens for cheap.
reply