bhouston

I bet it has multiple times but as part of white hate security testing.

This is science fiction, these models don't have access to their own weights

A bet a worm could pull along a 1GB file with weights in it and run it on a compromised machine, but luckily for us for now, 1GB isn't really enough to be really smart, yet.

We are sort of lucky that AIs right now require so much specialized compute+weight storage that we can easily "unplug" them remotely when they misbehave.

I wonder if that will always be something we can do? If they could bring their own compute/weights with them, or somehow tap compute/storage in non-obvious ways, we would be much more screwed.

Whether it is "open" or not seems to be in question. While it was initially called an "open" model, it seems that "open" mentions have been scrubbed from website.

When your customers are government mandated, are they really customers or hostages?

With the caveat that this is a biased sample. The main outlier is Linux with < 20% availability.

I use it from the Cursor harness.

Every time I use Composer 2.5 I have to spend a bunch of time cleaning up its mistakes. It is unusable compared to GPT 5.4 or 5.5.

My time is more valuable that I will use a model that doesn’t f** up my code base.

Passwords are definitely not better than passkeys for login flow as a user, at least not on MacOS / iPhone, etc. I just tap my finger onto the fingerprint reader on my MacBook or I just look at the camera on my iPhone - biometrics for the win.

But you need something as a backup to passkeys so I choose emailed OTP, but it isn't meant to be the primary way to log in.

Congrats BetterAuth! It was the system I was considering before I rolled my own auth system around the passwordless concepts of: OTP + Passkeys + Google login. It is quite nice and simple and I've ported it to 3 separate projects now just via LLM:

Talk: https://ben3d.ca/blog/passwordless-login-system

Live Demo: https://passwordless.ben3d.ca

Demo Code: https://github.com/bhouston/passwordless

If you are building a user system with a database already, adding passwordless auth is easy.

prohibition doesn't work in any of its many forms.

It definitely does work in that it will prevent us customers from accessing cheap goods from China and thus enrich the local AI companies that are paying the politicians to ban the Chinese models.

So this means that China’s low cost models won’t benefit the rest of the world? Then we have to make our own. If we only have expensive models and they have cheap ones they will be at a huge advantage not unlike low cost of labor advantages.

I think that token usage by engineers continues to increase, probably at a very high rate for many years (we are in the middle of the S curve of adoption and it isn’t yet clear where this will plateaux) but an increasing percentage of those tokens are cheap, because we use expensive models for goals and design and cheap models for implementations and workflows.

There is one in the EcoBee Premium and we use it to automatically drive our HRV (heat recovery ventilation.).

It is better to have it in the HVAC system than in your phone anyhow:

https://ben3d.ca/blog/upgrading-hvac-control

For any intelligence agency, they can afford to keep and store all of that forever, and later do analysis on it.

At the scale the AI companies are operating at, I think it isn't likely that they are sucking it all in right now.

More likely I think the intelligence agencies will get a real-time live tap into the raw data feed which they will process onsite for interesting things and then if things are flagged, they will log it in the intelligence agency systems.

All remote AI are a massive security risk for individuals/companies/governments that may be targeted by the US government.

It is likely that the US will get a live feed from each AI provider that they are inspecting in real time to identity things of interest, terrorist attacks or foreign government planning or even foreign companies competitive to key US companies.

It will give them access to the though process in those companies as well as much of their text-based IP (source code, docs, meeting transcripts, etc)

Also if you are using local AI that you didn’t train yourself you can never be sure it doesn’t have purposeful biases in its reasoning that may disadvantage you - such as directing you away from certain plans or ideas or patents etc.

Disclaimer: I didn’t post this out of support for Thiel but more just because it is weird a tech titan is making such claims.

He did the same thing with VR/Oculus/Reality Labs. Huge budget but it doesn't justify itself in terms of outcomes. Strange. Is it something in the way he/Meta manages the projects?

I think Meta has a problem that very few of its internal projects widely succeed. Facebook and Messenger being the only two. The rest of the successes were acquisitions.

This is crazy. Horrible OpSec and may end up getting people harmed or getting put in jail.

I do recommend turning off "isolate" for as much of your code base as possible when it makes sense. And I recommend ensuring "maxWorkers" is being used properly, I prefer something like 60% of my totals cores as the number of workers to use. And use a top level vitest start so it properly runs all the packages in a pipeline rather than as separate vitest runs (which would mess up the maxWorkers optimization anyhow.)

I have my 3000 test project suite completing in 15 seconds on my MacBook Air M3. It is pretty sweet with that setup.

Have you tried contributing a system for that?

Version 3 => 4 => 5 => 6 => 7 => 8.

I did the migrations and they were not horrible. A few breaking changes but they were relatively isolated. And the speed and improvements over these versions had been quite significant.

But I have zero clue what those other tools are.

The incorporated tools are actually really amazing:

- vitest, an ultra fast test runner. After using a lot of others, including jest and node's built in one, I love vitest.

- oxlint, replaces eslint but is compatible with its file format and ultra fast, since it isn't written JavaScript. I tried biome, but I found oxlint to have more rules and the eslint compatibility was nice.

- oxfmt, replaces prettier and is faster since it isn't written in JavaScript.

- rolldown, replaces rollup and is compatible with it but it is much faster since it isn't written in JavaScript.

The above are my go-tos in new projects anyhow since they are killer good and fast.

The fact that political spending is considered free speech in the US gives the rich and already powerful massive sway in the political system, it is basically tiled hard towards them. And then combined with how PACs hide their funding behind names like "Everyday Americans Making the World Better" when really it just wants to lessen online gambling laws for a billion dollar company, is just brutal. US politics is a dystopian future realized.

We need a vaccine for Lyme disease, it would be a lot more effective than paying $3K per kilometre of path per year.

You can not buy wifi usb adapters that work with MacOS on MacBooks I believe. I could be wrong but I think none of them have driver support.

Yeah, given the through put on the Pico is just 4Mbps in this setup, it will not work as a travel router for the family.

You could use the hot lines from RBG as the binary id (2^3) of the monitor and then average up the RBG intensity from the hot lines as the result. This would give you 8 monitor support.

Or you could limit it to 3 monitors but two hot lines pick the monitor and then you can sum the intensity on those two hot lines two double your fidelity.

If you do the two line color system you could have a YC signal of luminance (Y) and chrominance (C) and thus you have a color signal.

Interesting project.

In a similar but opposite vein, I am going on a vacation and I wanted to share the stupidly expensive internet in my room at night with the family so I am likely bringing a raspberry pi to have as a travel router attached to my Mac. In this case, I can use the RaspAP project: https://raspap.com/

This is slightly different in that I do want a NAT.

Inflation numbers aren’t even believable and It’s already not fun

For inflation to have an impact on the US debt, it has to be approaching the level at which the US debt is increasing. In the last year, the US debt increased by 7.6%, much higher than inflation.