greesil

How do you know this? I'm not trying to attack your statement, I am genuinely curious how anyone knows anything about model performance outside of benchmarks that are already in the training set.

Yes, I understand one function of jargon, which can be useful to insiders in that it conveys a precise meaning. But, it can be confusing to outsiders, and that is also a useful thing for insiders. In the context of LLMs, what other function can produce p(next token) if not the LLM? And, you just about make my point for me with regards to jargon being confusing by misidentifying what the policy actually is (something i never would have noticed :) In any case, it's an interesting paper. Thanks for all your down votes everyone.

But the probability vector is the output of the LLM, no?

It's a comment. On Hacker News. Not the RL subreddit, or whatever. I'm just amazed at the jargon. I'm sure it's useful, but one could just call it model output.

Wtf is a policy? Is this some sort of RL thing that I'm too ML to understand?

Gemini tells me it's the probability of the next token for an LLM. Okay then.

"This is notably fast given that this is the first time that an Android driver bug I reported was patched within 90 days of the vendor first learning about the vulnerability."

This makes me feel better about Google, but also makes me kind of frightened of the rest of Android. I wonder what Apple's response time is?

Which comments?

The whole country? Really?

Some people respond to incentives. The rest of us are just trying to do our jobs and will probably be fired and then later consumed by the basilisk. We are living in an age of extremophiles.

Too bad they can't just easily vibe code new features.

Isn't this what RAG is really for?

Well this explains tripping on acid a little bit

Insurance company with an army primarily. But now with an expensive credit card bill.

You did not increase comprehension by not using a single negative.

How do you keep it from getting prompt injected?

Oh I get it the runtimes are nice and small, you're using Claude for the intelligence. Obv

I think I'm just impressed with anthropic more than anything. Defcon would have me believe that prompt injections are trivial

Which company?

Suicide booths a la Futurama

Collective action is not our strong suit.

That's why I'm still using Tomcat and web cgi

Lol. That's cool, I've never heard of that!

I was thinking something like an esp32 + mesthastic / LoRa + REST API on the LAN, discoverable via multi cast. The "landline" is a tablet or phone with an app that talks to the esp32. Separately, a parent with the app does the Diffie-Hellman key exchange over SMS, NFC, or some other channel with the friends who also have the app, and you know their identity. The phone app updates the device with the friend's keys, they do the same thing on their end, and voila you're in business. The kids can talk securely, you can read that the kids say via the LAN, no goddamned third parties.

It must have changed in the last 8 months

I was thinking of doing something like this for text with LoRa. But, having kids I don't have time to do that. This seems really great!

I read the previous discussion, oof:

S04dKHzrKT wrote

Make note of the privacy policy[1]. Some users may not like the data they collect. > Information Collected from Children: As detailed in Section 3.C, we collect voice audio during calls, call log information, and utilize the Parent-provided contact list in relation to the Child's use of the Tin Can Device. We may also collect device identifiers and technical usage data related to the Service.

[1]: https://tincan.kids/policies/privacy-policy

Har har

Excellent point. Maybe it's the goal of this attack to demonstrate this capability.

I'm not saying 60 minutes is a bastion of journalistic integrity before or after Bari, but this episode was relevant and interesting https://youtu.be/C1jmAj9OUOs?si=EyFHpPRJ9OF36Xcn

I can appreciate boring nowadays.

Do you like things that look like bank runs? If so, then yes.

How would you know?