gleenn

Distilling isn't necessarily easy, there is a huge cottage industry of services middle-manning ChatGPT and Claude to collect huge amounts of data. It is still vastly cheaper than training yourself, but it is certainly not easy or feasible for most organizations. And I'm sure a flock of lawyers would show up if someone in America was found doing it.

That particular quote did stand out to me. I think the counter claim is that if everyone uses software everywhere for everything, there is a very real societal cost for damaged data and if it's hard to detect then there is no way to prevent or measure harm. I think the "invisible hand" free market is showing some serious strain in modern times. People stopped using leaded-gas because it was causing measurable decreases in IQ for generations. I don't care how much you want to save a few dollars if it harms a lot of people and is difficult to measure. Everything else seems like a race to the bottom price so pushing the market in this manner doesn't bother me in the least. If you are running a server with others' data at stake, it seems like an obvious benefit for society on the whole to address the problem rather than serve some cheap datacenter operators.

One of the Mars landers famously failed due to unit conversion errors from metric to standard.

If someone quits their job, do all their opinions suddenly become suspect? You're kind of damned-if-you-do-damned-if-you-don't. Either you work for the company and you are biased one way, or you quit and now your bias is now suddenly the other way. I've joined and quit many jobs and my opinion may or may not have changed due to my change in status but it is clearly and ad hominem attack.

Listening and responding are likely pretty different mental activities

I think the fundamental problem is that training current SOTA AI models is very expensive. If a simple "classical" model can detect them, presumably at much lower algorithmic cost, then why wouldn't the model trainers use these same tools to feed back into their models to improve them at low cost to make them better? It's an arms race. Any cheap pattern can and presumably will be used to retrain if it becomes and effective way to catch AI.

I think one thing to point out is "everyone" in this context is probably the broader market, but the stock holdings aren't distributed uniformly. A few large investors could actually be sophisticated and unload a fortune and the rest of the shareholders could still largely still ignore fundamentals and suffer large losses. The broad market can be ignorant and the stock can (and is) be down, both things can be true.

"Tilting the windmill" is from Don Qixote who famously fights a windmill (which is a bit absurd) so he is courageous but also a bit foolish.

Extra props for tilting thw windmill that is tech behemoths funneling data to government agencies without oversight. Aiming at Amazon is certainly something not to be taken lightly.

Correct, AI will not replathe 3 high-paying watch maker jobs that exist. You are the best kind of correct, technically. But you are distracting from the fact that most people aren't doing anything even remotely physical related in the space that some people posit will be decimated by AI: white-collar jobs where you are a keyboard jockey all day.

If you are using a mechanical watch to keep time (accurately for long) you are doing it wrong

A major upside to Costco is you can actually see stuff and you also can walk out of the store same-day. Also I never ever worry about the counterfeit and/or low-quality crap you inevitably get from Amazon. And if Costco sells me something crappy, I drag it back in and don't even have to start up the printer (it's a zombie at this point). Costco has a running rule that they never charge above 10% in profits so I know I'm getting a good deal too.

I heard someone argue that because original phones rolled out in the US much earlier and more broadly, we spend all this money on copper wiring. Because Europe didn't have as much, they transitioned to fast cellular quickly. Being first does not mean you will be best certainly.

Whether or not you find Anthropic's behavior bad, theybhave been very loudly stating the foreign labs have been distilling their models for a while now. This seems like an obvious response to me that would be a mechanism to make that obvious.

Sounds like a case of the Mondays, Mike Bolton

And the fact that their system doesn't dump water. I think that is actually perhaps the bigger deal. Datacenters have been getting a lot of heat (pun intended) for using significant fresh water at the expense of local municipalities.

A Mac laptop can be had with 32GB of RAM for far less than $4500. Not sure if they actually need 32GB of discreet GPU RAM. My Mac laptop does run Qwen at a reasonable speed.

Shouldn't you care more about the actual issue than who is writing the laws around it? Why are you so pissed off about the "who" instead of happy you are getting what you want?

It exists, although people complain it is too noisy. You can hook in any if your own tools too.

https://github.blog/security/how-to-scan-for-vulnerabilities...

I had the privilege of working and sleeping in the original Craigslist office/house in San Francisco. It was just another typical, ageing house they had rearranged a little to have a ton of deskspace in the main area. A lot of start ups (including Zappos IIRC) had also been there over the years. They had a mattress in the loft/attic you could crash on if you were up late too.

Claude Code regularly installs dependencies using (p)npm after I e.g. pull a company main branch to get in sync with my teammates. That happens often. So I pull, Claude edits some code as you requested and it should pass because Claude did alright, but your local box has out-of-date deps. So then Claude runs (p)npm i and now we have automatic exploitation of this gaping hole in npm given extremely common and current AI tooling. Someone has to figure out how to stop AI from running that command or NPM needs to stop that behavior, and I guarantee you it will be easier to get one tool to change than all AI.

Friends don't let friends use NPM. At this point it is so wildly crazy watching people get owned, I don't understand how anyone uses it when they could use e.g. PNMPM and block one if the most obvious and frequently exploited holes. These tools with arbitrary code execution when trying to download some code have got to stop.

Edit: typos

I haven't heard anyone specifically state their justification for blocking bio research along I can only assume it's to prevent manufacturing bio weapons or virii?

I personally find OOP to be the most intuitive for large scale systems design, but that's just me.

The beauty of Clojure shines through when you want to change something that cuts through a large part of a large project. If you are using mutable data, you may end up with many bugs from various pieces of code mutating objects inconsistently. With Clojure, if someone hands you data, you can't possibly break some distant piece of code by updating an object: it's just not possible because you only ever make fast, updated copies. The more complicated your codrbase gets, the more this benefit is realized.

I actually kind of think of it as an easier mechanism with similar outcomes to Rust's borrow checker. Only one piece of code ever owns the data so things end up much safer. However it is way easier to use IMHO because you just know that zero people own anything and everyone can read everything.

It also makes converting some code to be multi-threaded extremely easily and with some constraints guaranteeably correct.

Lots of dovetailing features neatly put together for both clarity and less bugs and more usable cores which are probably sitting idle.

Because they are breaking all their own rules by removing the seasoning and profitability requirements to fast-track this stock in.

Not if the reason is "I want to stalk my ex".

Sometimes I feel like it's a cover for some other org actually just wanting to steal the data and this being the excuse.

Technically it's a transpiler.

That has definitely not been my experience as of late. I have produced multiple, largeish Clojure projects with AI that have been perfectly formatted and functional. Perhaps you were using an older or possibly smaller model? I am admittedly using Claude with higher end models and mid to high effort but it has been working great for months for me at this point.

It's surprising to me this is news. Governments buy and install this equipment and it flags license plates and anyone thought that wouldn't be used for things like immigration control? I'm not saying it's right, just that it's shocking people wouldn't realize that.