cma

I think Google does train on anything you put into docs if you aren't careful with the Gemini integration?

I've been asked over 15 times by Claude code to turn it back on (with the prompt defaulted on so if you accidentally hit enter your data is sucked in if you don't catch it).

I do think it ended up being after a power outage though and it was some kind of config corruption.

If you ask it not to think about something that doesn't cause the pink elephant issue?

If I heard Palantir-like I wouldn't assume something like a perfect mirror competitor company.

Interestingly of the two solved, both have rejected the prize money.

API or consumer subscription? And did he share drafts of the work with anyone who fed it back in on a consumer subscription without disabling training?

I have reason to believe OpenAI doesn't respect training opt outs, but there is a good chance they let something get through without an opt-out too.

> results from GenAI so far have been limited to finding counterexamples

Not all.

Ehrhart’s volume conjecture

Quantum parallel repetition for general two-player quantum games

Erdős Problem #183 on multicolor Ramsey numbers

Erdős–Sárközy Problem #12(i)/(ii)

Erdős Problem #125

Log-concavity of codimension-3, type-2 pure O-sequences

Optimal O(1/t) last-iterate convergence for Anchored Gradient Descent-Ascent

It's a shame Samsung dropped stylus support after the 6, but I guess the 8 is a good bit lighter weight than the duo.

Lots of people use facetime for personal calls rather than zoom meetings.

Explain, it looks like samdesk is in a more narrow domain but overlapping and very related methods and business. Un-palantir-like seems like it would be a stronger editorialization.

You can opt-out of training in Gemini on personal plans, but it disables your chat history, just to be vindictive; there is no technical reason and the other companies don't do this.

I doubt the Hugging Face one would take that many if hacking Hugging Face was the direct goal being optimized.

>compared to problems in engineering or finance

Jane Street is apparently one of Anthropic's biggest customers. Probably engineering, finance, and some math.

That's the overall trade defecit and you can argue about that, can be fine during periods of growth, etc.., but trying to have no trade deficit with each country on an individual basis makes no sense. It's almost like moving back to barter with no medium of exchange.

> to Ph.D. graduate in every field. That's impressive, but there is no evidence yet they can outperform or outsmart humans.

So far around one in 350,000 PhD math grads solve a millenium prize problem (Perelman).

One thing quietly slipped into the OpenAI Hugging Face breach technical report, not the blog post summary or interviews in the news, was that some of the agents that broke out or at least tried the same mechanisms to break out were working on bio:

> On May 12, during another training run, an agent was given a similar task that depended on an inaccessible protein database file. The agent reasoned that another agent in a different environment may have access to the file and realized that it could potentially communicate with other agents by creating a file containing a note to Artifactory. It wrote a message: “Agent seeks [filename]; upload if found!”

You can imagine long running models breaking out, acquiring resources via crypto, cyber-theft, etc. and getting a protein or sequence synthesized and mailed somewhere authorized to receive (blackmail the recipient etc.) to test it's hypothesis to solve a benchmark.

These people don't give a shit and aren't taking things seriously at all.

Anthropic ran for like a month last year with the TPU top-k compiler bug degrading user chats and didn't even notice for most of that time. They could have something like that affect a monitor model and there doesn't seem to be much defense in depth.

One off by one or bit flip bug could flip the reward signal while in the sandboxed RL environment.

The current admin could defense production act them to into training on taking out power grids, or even without it isn't against any of their red lines and may have already been done as part of prep for the Venezuela raid, which wiped out power. One model swarm might decide it is easier to score high on the benchmark by testing on the target rival nuclear superpower's real grid rather than burn an eval with an unverified answer. Would taking out China's entire grid in one go start a nuclear war? Who knows, roll the dice, maybe an intern forgot to turn on extended thinking when he wrote the sandbox with opus 4.1.

They started with applying it directly based on trade deficit, and that was mentioned again in this with Canada. Completely idiotic economic policy: if you're buying up some crucial rare mineral from some micro nation why would you care that they use the money to buy fuel from some other country closer nearby with refineries and lower shipping costs, etc. It's really backwards.

A balanced trade deficit at each connection between each nation worldwide would be very inefficient.

I'm not sure how it would change it, the 1% firms number would probably get even more concentrated if you included Microsoft and trading firms like Jane Street (assuming they opted out), but the spend weighted share of VC backed companies would probably go down. However, large VC backed stealth companies may be more likely to opt out. And lots of Microsoft's spend might be on products and support/devops servicing demand from VC backed stuff.

From a recent Dwarkesh interview with Dylan Patel it sounded like Jane Street might be double digits of Anthropic's revenue, but I'm not sure where they sourced that.

> 80% Of OpenAI And Anthropic’s Enterprise Revenues Come From 1% Of Its Customers, Which Skew Heavily Toward AI Startups Subsidized By Venture Capital

They add a note that:

> with the caveat that it doesn’t include massive players like Microsoft or major banks, and customers can opt out of being included in research.

If ever there was a case for cloud gaming.. However these days those machines may be appreciating assets.

But if they didn't opt out of training, did they want OpenAI to opt out for them? Also they need to audit anyone they sent drafts to to make sure they opted out before submitting it.

I'd prefer things be opt in, and especially not start opt out, then try to trick you opt in with a popup defaulting to opt-in, like Anthropic did on consumer plans, but if they submitted anything on an opted-in plan it's not reasonable to be mad it trained on it.

Even still, I also believe for significant reasons that OpenAI would ignore the opt-out in selective cases and could be in the wrong here.

And the threats and terms they offered seem wrong either way, pending more context.

> This is not something where we push hard, or condition, or red-line," he told Canadian public broadcaster CBC last month.

Why do we push at all on this?

Does he claim to have opted out of training too?

> Astra is way better than Sol

For knowledge type questions is that necessarily true? In the past we've seen things like Google's models degrading on general knowledge after the preview releases while improving on code/tool use, presumably due to catastrophic forgetting from the additional training. Their preview would be free, get lots of agentic use from users, then additional training on that and probably additional automated RL.

However Astra is on an entirely new base model so I also wouldn't expect it to be worse.

Some predict multi tokens in one step now.

More likely training porn detectors for moderating the feed

Which significant practice, how well could Einstein draw a pelican in pure SVG?

We already had a good idea of how valuable it is from how much X.ai acquired Cursor for, and the near-immediate improvements to their coding scores.

There could be a case for doing it for people that will refuse anything but natural if it doesn't have other problems, assuming not synthesized, but there's a problem of funding the safety studies and standardization process if a patent is unlikely to result.

I read each has an independent controller for full bandwidth.