Meleagris

Members of the Senate committee reviewing C-22:

Marty Deacon (Ontario) Committee chair 613-943-3429 Tony Ince (Nova Scotia) Deputy chair 343-991-5540 Andrew Cardozo (Ontario) Committee member 613-996-9545 Katherine Hay (Ontario) Committee member 343-991-5536 Rebecca Patterson (Ontario) Committee member 613-995-2185 Hassan Yussuff (Ontario) Committee member 613-944-1108 Kim Pate (Ontario) Committee member 613-995-9220 John M. McNair (New Brunswick) Committee member 613-995-5060 Victor Boudreau (New Brunswick) Committee member 613-947-4110 Judy A. White (Newfoundland and Labrador) Committee member 613-996-2433

This is a cool article to come across, having just finished Becky Chambers’s new book, “As You Wake, Break the Shell”.

I think the advantage is making the honeypot goal-directed.

High-interaction honeypots are about extracting as much information from the attacker as possible.

Instead of just observing, you could give the model intelligence objectives and have it dynamically steer interactions to achieve them.

Deliberately leading an attacker down paths that reveal their capabilities, tradecraft, or infrastructure.

You can do this today…but it requires effort to develop the scenarios. With this, it’s just a disposable goal.

This is more useful than you’d think. Consider using something like this to drive a high-interaction honeypot.

I’ll be interested to see how this compares to Nvidia’s Nemotron 3 Diarization model for diarization. For my use-cases real-time diarization is still not where I’d like it to be, but nemotron was definitely a step above the previous attempts.

Edit: apparently it’s not open unlike the previous, so not useful for local. Sigh.

Ok. But why would anyone buy a third level domain when they could have bought the second level? It’s a strange thing.

Were the third level domains cheaper?

Me personally, it wouldn’t even occur to me to register a third level directly, because why bother when you can have the second level and do whatever you want with it.

Super strange.

Holy smokes!!! It has just occurred to me that anyone who can type on a keyboard knows how to play an instrument. Your keyboard is the instrument.

With solo mode, I asked ChatGPT to translate the songs to keyboard notes, and I was just typing the notes as if I was doing a typing test, and I was playing songs. lol the easiest "sight-reading" I've ever done.

Is this the every persons instrument?

Here is Happy Birthday in solo mode:

E E T E I U E E T E P I E E B X I U T C C X I P I

Crazy unlock!

INB4 "OSS Rewrite"

Sure frontier models can continuously scan projects for vulnerabilities, but projects written in memory-unsafe languages (i.e. C) require an endless cycle of discovery, triage, and remediation as new features are built.

Just wait a couple more years, I am betting that it will be economically preferable to migrate software onto memory-safe languages and more verifiable architectures rather than repeatedly auditing systems built on fragile code.

If they truly wanted to do good, an "OSS Rewrite" service is a natural logical successor to vulnerability scanning. Instead of spinning wheels and burning tokens finding bugs in unsafe code, it would probably make more sense for AI to help move projects onto platforms where entire classes of vulnerabilities are just eliminated.

Totally agree with this article. With AI, I think now is the perfect time to consider where Unikernels might fit into your architecture, and how you can leverage them to minimize your attack surface.

I’ve started looking at them myself, and have been comparing them to mature VMMs like Firecracker, and asking myself where each piece of my stack might be best run.

Virtual machines and containers are no longer an effective isolation mechanism when AI is involved. VM breakouts are becoming trivial. So everyone should be considering how to bake better security into the their runtimes.

I’m surprised people don’t get refusals running this, or are they running it with cyber-enabled models like daybreak?

In my experience, even just hinting at reverse engineering to models from Anthropic or OpenAI leaves them extremely sensitive to refusals. After all, the same techniques used here can be used to find exploits.

Or are people using it with local models?

Curious.

I’m actually more interested in hair removal. There should be better methods than what we have now.

If I’m going bald, then I’d love to shortcut the process.

Shaving is awful no matter where it is on your body. It takes time, it’s irritating and you never look fresh for more than a day.

I wasn't expecting to get hit by AI generated slop on Standford's website. At least mark the article as being produced by AI.

> “What they found was striking: … These differences essentially locked each progenitor cell into its respective fate, like travelers on parallel tracks that never cross.”

The dramatic colon, the obligatory closing metaphor, makes me ill.

So I did test this, and it doesn't work because the quantized layers need torch._int_mm, which PyTorch's MPS backend doesn't implement. It just throws NotImplementedError.

Anyone who is bothered by mosquitos should look into how to build a “bucket of doom”.

Basically you deliberately create the best mosquito breeding habitat in your yard (stagnant water), then spike it with a a bacterial larvicide to kill the larvae at the source.

Since mosquitoes don’t travel long distances, you end up creating a small mosquito free zone around your trap where they were never able to breed. It’s extremely effective