0xbadcafebee

The first extension I add to any coding agent is one that redacts or rejects any read/write of a secret or credential or credential file. It's annoying, and the agent can work around it, but better to add a weak guardrail than none at all. Best is to not give it access to any secrets, but that's not always possible.

Here's a few for Pi: https://pi.dev/packages/@k3_2o/pi-secrets https://pi.dev/packages/@spences10/pi-redact https://pi.dev/packages/keyshroud

A couple for OpenCode: https://github.com/boxpositron/envsitter-guard https://github.com/kenryu42/cc-safety-net https://github.com/inkdust2021/opencode-vibeguard

If you don't let it read prompt-injected text, it can't get prompt injected. Don't give it a tool to download web pages or read files (other than the ones you control).

> someone drops an update into your adblock software that writes out a log file where the text reads ...

They don't need to prompt-inject an AI if they are already dropping custom rigged software onto your computer.

Ruthlessly optimizing isn't the best goal, because people "optimize" all the quality out, and lack of quality leads to lack of sales and increased cost. Ford/GM may have "optimized" their line for efficiency, but the increase in defects caused more losses. Companies that practice Lean tend to forget this and focus only on speed and cost, leading to enshittification. TPS focuses more on quality and its effects on lowering cost and increasing sales.

Toyota is also a good example of what not to do. Their lack of earlier research investment in EVs has led to being severely behind the market, in addition to their lack of focus on quality causing poorer sales (their quality used to be one of the only reasons to buy them). KIA basically stole their playbook by creating loads of good-quality cheap hybrids and EVs. Toyota could've had that market if the invested more ahead of time in research.

And that's what's happening with AI. While there are plenty of players now, it's the ones that aren't looking ahead that will be left behind. Nothing scales forever; when you hit the ceiling, you're trapped in a box of your own making.

In the short term. In the long term, his research on learning will be the next revolution.

Executives chasing short-term returns. Google did 20 years of AI research, even creating custom chips, just so we could get to this point. Without long-term strategy Microsoft will perpetually be the also-ran. (which I guess is fine with shareholders?)

That's kind of the point of a consumer society; the people's wealth is increasingly consumed until there is no more left.

Remember how tiny the drone market was, until DJI made them juuuuuust cheap and useful enough? Then suddenly there was a drone market, and DJI made a better one, and then the market got bigger, and then DJI made a better one, and then the market got bigger, ....

It is logistically impossible for the West to manufacture robots anywhere approaching the scale and cost of China. They control the materials, the refinement, the advanced manufacturing, the vertical integration, and a cheap labor force, to say nothing of a planned economy that can pivot on a dime to execute any demand there's available capital for. It is laughable to think the West can produce robots that compete with China.

If tomorrow morning we decided to pull a Manhattan Project just to match China's capacity for cheap effective robotics, it would take 10 years. By then China has already found their product-market fit and secured the world as a customer.

Over the years I've found so many Fortune 500's with either massive bugs that broke the site for customers, or huge security holes, and spent hours trying to find a way to report it to them, to no avail. The people who make these websites just don't give a shit. So I now just look up WHOIS and email the domain's contacts and give up.

It serves a useful function in the same way that a bar serving alcohol serves a useful function. It's making money by giving people something that's bad for them. They really want it, so it's not going away. But it's a net negative on society and we'd all be better off if they disappeared.

Some gas engines have low end torque, big block V8s and anything turbo. But yeah if hauling uphill is your thing diesel is a no brainer. It would be great if our half ton and midsize trucks had diesel options but as usual other countries have better options.

Personally I'm going hybrid truck next time. The advantages of electric (torque, 240V inverter) with the flexibility of gas. Payload is lower but I'm hoping they fix that by the time I need a new truck

"News" is just entertainment and fear mongering for money. Always has been. There is no other point to reading the news. Or as someone else put it: "Newspapers are the original doom scrolling".

It would be nice to have a populus informed about what's going on in local government and actually understand the issues. But nobody really wants to be informed. They want biased ragebait.

Journalism pretends like it's fair and balanced while simultaneously only reporting part of a story that's much longer than presented, in order to provide the reader a specific perception. In other words: manipulation. Ignorance is a better outcome, because then at least people aren't overconfident in their wrongness.

I might use them if their search results were better. But DDG's results are the same and free.

Personal use: $10 subscriptions through Charm Hyper and Synthetic, partly to try them out, partly for reliability when one acts up or is slow. Pi for an agent, CLI in Tmux. Dingles for project task management. Linux. Was using pi-web-ui for UX but it's a bit too crap. Handy for stt.

Work use: Claude Enterprise and an OpenCode Go subscription (for when Claude is down). Termic for UX to switch between parallel agents running both Claude and Pi. Linear, GitHub, Atlassian MCPs/CLIs. Custom tools written to mint read-only credentials for headless sandboxed AI agent work on tickets. MacOS. FluidVoice for stt.

A mix of Pi extensions for work and personal. Personal is a bit more loosey goosey, Work has more explicit gates and limits. Customized browser agent, computer use, Matt Pocock's skills. 1Password for loading credentials as well as a custom CLI/GUI to prompt me for input when needed.

> letting badly programmed software die when unexpected things happen

Go ahead and program a flight controller this way, I'm sure nothing bad will happen.

Musl has a history of introducing (and not fixing) major bugs in their library that aren't in glibc; I would not depend on it for production

The orders of magnitude fewer security vulns for OpenBSD are not due to less testing. Everybody has always known OpenBSD is more secure, and it's not from using a different language. The reason is the code. Linux was not designed to be secure. OpenBSD was. It's that simple.

Funny how OpenBSD is "memory-unsafe" yet doesn't have these issues. Could it be that the programmer might have something to do with it?

It's not technical debt. It's lack of engineering rigor.

Bridges don't get driven on until they're inspected. And they only pass inspection if they were built according to standards, and meet design specifications that have to take safety into account.

How we get there, how we enforce it, is another conversation. The point is that we have to start getting serious and be real engineers. The bare minimum standard should be that you can drive a truck over it without worry. How much software can you confidently say that about?

$3,000/month is nuts! For $36k (1 yr of spend) you can buy a system with dual rtx pro 5000 (144GB VRAM). or spend a little more ($44k) and get two M5 Ultras w/512GB RAM. the latter, clustered, is big enough to run the biggest open weights (quantized), or smaller MoE weights on each in parallel. what you get for these systems is way more than $3k/mo of tokens, because you can run it 24/7/365. you can have pass after pass after pass of optimization, bug fixing, orchestrated subagents on backlogs, etc. way more value than paying for managed inference

AWD provides more traction while accelerating, which makes it safer, which is why people use it in winter... Saying it's less safe is ridiculous.

Now if you happen to be really good at driving, it's even safer because you can recover from a slide faster, again because of the increased traction during acceleration. Once your tires have traction again the additional traction moves you out of the slide faster.

The only problem here is people going too fast which can happen with any car.

It literally just did math on some numbers based on a set of rules provided in the prompt in order to sketch a visual of a game. It didn't "interpret model weights" any more than an MP3 player "interprets physics of electromagnetic waves". The LLM built its own decoder on the fly for the games... which is what it's doing when you ask it to write code, calculate math, estimate the structural load bearing of a beam, etc.

LLMs are extremely hyped up right now. It's like when everyone was crypto-crazy, and threads were full of people who had no idea what was going on but were very excited.

It's not really exaggeration. It's doing something similar to abliteration, but more. First it finds the part of the weights that does a thing. But then the difference is, while abliteration deletes stuff, this is adding/changing things. It does this by doing semantic analysis of the weights, changing a little bit, see what happens. It's not super complex or advanced; it's essentially a form of statistical analysis and experiment.

This means we might not need to fine-tune with thousands of passes and tons of GPU. We could just tell a frontier model to edit some weights and iterate until there's better results. This could cut the cost of fine-tunes by 100x. Makes no difference to the normal end-user, but pretty huge improvement for LLM research and model tuning.

It's less an industry thing and more a war thing. US wants to fight more wars and for that you need a plentiful supply of portable energy that other countries can't block. For physical wars that's petroleum fuels. By controlling "Greater North America" they can secure their cheap fuel supply and hunker down here. Investing in green energy would de-emphasize the infrastructure you need to process and use massive amounts of oil for cheap.

The coordination is organic. Simply start with this assumption: "each device will check [something] every 60 seconds". Within a few years, all systems will be synced.

When the power goes out on the grid, everything is turned off at the same time. When the power comes back on, everything turns back on at the same time. Every machine that had a 60 second timer is now operating from the same starting time.

That ship sailed a long time ago. Science decided that killing mice for experiments (regardless of the experiment) is fine. In fact, a large amount of the mice killed are just redundant (not needed for a particular experiment)

In computer/systems/distributed engineering this is called request smoothing or traffic shaping, and it's implemented by adding jitter, which is a random time offset/pad added to the interval.

The funny thing is, modern electrical codes already account for this phenomenon. In order to implement Net Metering (selling your power to the grid), modern standards implement two kinds of jitter:

- Sandia Frequency Shift: a minute shift in the power frequency generated by inverters. If the grid is connected, this is effectively eliminated; if the inverter is disconnected from the grid, it detects the frequency has shifted and disconnects the inverter from the grid.

- Slip-Mode Frequency Shift: basically the same thing, but phase shift (which I don't understand).

Both of these are used for anti-islanding (don't create an island of power). When the inverter detects it's isolated and stops outputting current, it then has to check every 5 minutes, with jitter, for whether the grid is restored or not. It has to add random jitter, because otherwise all the mini battery/solar installs will come online at the same time and swamp the grid, causing another outage.

The tech will never be there with LLMs. They can't learn, they hallucinate, lie, cheat, steal, destroy things, break the law, etc. You can't train them out of these behaviors, it's how LLMs work.

An LLM is just a really fancy nail gun to replace hammers. Much faster nailing, but you need to hold it right to not shoot yourself in the foot.

Some of them might, but there's no way to know. The 1% that might be experts look and sound exactly like the 99% that aren't.