stefan_

The open internet is now a cesspit, with very little new good data. Expect everyone to train on user data always. They just got clever about whitening it.

Actually they just Arc<T> everything

"I don't know anything but I don't like Graphene"

Well we know Google isn't enabling MTE, while LLM-enabled exploits are multiplying rapidly. Maybe you need to get back to work?

The US just added a colony in Venezuela in cooperation with their local Reichskommissar and is looking to secure its oil. American perception of their country is quite something.

The systems jamming in Hormuz and Ukraine aren't exactly cheap and low powered, nor are those used by the US military in one of their many jamming exercises every year. Cheap GPS jammers fit in a cigarette lighter and are used by truckers who don't want their boss snooping on them, but they don't quite threaten aviation.

That aside, you don't need to give up on GPS just yet. As expensive as airplanes are, there is no reason not to have at least 8 element CRPA receivers in them. Mostly a military niche product in the past they are rapidly proliferating, and most fuselages make for a great ground plane.

I don't think this is something anyone who has ever read a "thinking trace" would unironically say. Not that you can even see them in Claude.

When thinking first started and you would still see the whole "thinking process", I thought it was a ploy to 10x token use because it was just the most inane bullshit. "But wait, the user is asking me to" in loops.

It's crazy how much they have enshittified credit cards. It's not stuff like CVC, that's fine, but all the random failures! Nothing will turn me off your "payment system" more than having it randomly fail on me for utterly incomprehensible reasons when I'm simply trying to pay. It's five times worse when you are traveling, buying some small items in a shop and have it fail at the register.

You can point AI at any AI produced code and ask it to review it, get back 10 bullet points and a few pages of prose. And the fun part is, you can do that over and over and over again!

I think these "Flash" models are sort of an evolutionary dead end. Sure, there are some routine tasks and applications where they can be used. But for the actual novel development work? It's much better to run a big model at high power for 30 mins than watch the Flash model struggle for 2 hours and produce massive churn.

Same reason your phone has a few big CPU cores for real work, it's much better to "race to idle" than have an "efficient" core struggle. Shitty experience, shitty power efficiency.

The juniors I have seen don't even try to understand anymore what they are building. They don't understand the product, they don't understand the technology. So everything AI is building for them is nonsense. "Reviewing" it is impossible. Some of them think they are being extra clever having Claude process review comments for them. Even if you hire one, this way they are out the door within the month.

They were also caught having their Grok code agent upload entire source code repositories in the background.

I guess the joke is "come to Origin, your code is already there".

That's mostly because at some point a wave of nonsense swept over the field that brought with it the Scrum master, agile, pairing, middle managers thinking up elaborate Git branching schemes (they don't understand Git) and of course, the mandatory code review.

It's best to take all these things in moderation.

In Poland fees are capped at 0.3% for credit card, 0.2% for debit per EU. Which is why generally in the EU you don't have the insanity with credit card loyalty programs.

I hope there is interesting research being done to stop models from 9B to 3T from authoritatively giving the wrong answers to factual questions instead of doing tool calls, but this article will tell you nothing about it. That's because it's clearly prompted by someone who simply noticed this deficiency very obvious to all practitioners and even users.

CLAUDE.md is mostly powerless against the reinforcement learned crap. I'm up to three separate instructions telling it to cut out the hyper verbose, retelling history comments and it still writes them every time.

There are really a ton of fields where people have done an absolutely terrible job of properly using the advantages of software, even pre AI. Like, even today there are people manually drawing footprints in PCB design software, manually determining what pins carry differential signals, manually forgetting to swap RX and TX lines.. except for the functions they absolutely had to have a computer do, they forgot to move out all the drudgery and footguns and continue operating in "PDF space". It's embarrassing for a profession that is supposedly made of engineers.

PDF space is still more advanced than the people typing into "word processors" however. Making software to emulate letters was such a massive mistake for the trajectory of computing, it let people effortlessly "adopt computers" by making a digital facsimile of their terrible analog processes. Only with AI we now have the ability to digest all this unstructured mess.

It's really getting extremely obnoxious. There is no point to the whole "shell commands output" view anymore because the model aggressively redirects everything to some scratchpad file, so besides the output viewer being a broken mess, it doesn't show anything regardless.

My favorite is when it wraps commands in convoluted pgrep wait patterns that end up matching it's own bullshit, and thus never exit. It also feels like Opus 5 will now always write some adhoc Python script to MAKE UPDATES TO FILES, which seems extremely terrible.

The complicated shell commands also constantly trip up against quoting rules, requiring another iteration.

This is just finally bringing the magical "the computer did it" to the masses!

When companies like Amazon accept, warehouse, bill for and ship unsafe Chinese trash, well, at every step it was just a computer. Google, X and others will happily show state funded scam advertisements from fake accounts and there is no one accountable, because it's just the computer handling those.

Now with AI, everyone can claim this magical legal immunity for themselves.

LiDAR in drones is not necessarily the "stereo vision" application you are thinking of, it can also be something as simple as a LiDAR altimeter.

Funny how that works, so are Anthropic and OpenAI apparently. Maybe there is something we are missing here!

Turns out you can't pay em as much as VC will, haha

Maybe you are just aren't aware that military GPS jamming is extremely routine around the US. It affects airplanes more than cars because the curvature of the earth can not help you and at typical flight altitudes you have easy line of sight to the jammer.

That's an odd claim. Human failures are uncorrelated, but the Waymos all running the same software must be expected to fail in the same way, in the situations in which they fail. Which is why they often make the news; one Waymo stuck soon sees many of its brethren join it.

Prices going up 1000%? Time for maintenance!

If you want to see their approach to maintenance while prices and needs are high, watch them put oil refineries back in operation days after drone hits.

And then Russia refused to install the turbine, proving it was a ploy all along. Like the maintenance in the first place. Which is why they acquired gas storage in Germany, then ran it low timed with their illegal war of aggression. This is a theme of course; Russia invented another gas crisis every few years with Ukraine. It should have been a warning sign when they settled them by taking payment in the form of Soviet-era cruise missiles.

The gas was never cheap, either; apart from the money spent now to restrain the military buildup and invasion it financed, even at the height of supplies it was much higher price than China is paying, a price sustained by Russia using military force and more to restrain other (Asian) countries from supplying to Europe.

(This was obvious to everyone, right? A single majority supplier nation state charging a cheap price out of the kindness of their heart? No, of course not; neither cheap nor kind. It was a monopoly, built on blood.)

Yet the README is generated ("Two skills maintain the boundary in both directions.") and the demo is .. a Mobius strip?

I'd prefer if you explain what it is you are building in the form of human-written text.

This is a technical problem with technical solutions. You can just fix things, you don't need to throw out 30% of your power generation to pay billions for climate change vagueposting.

Because 90% don't even do anything? It turns out it's actually one of those really annoying problems to delay cookies which were supposed to be sent already in the HTTP request response until a user interaction has happened. And on a lot of pages, non technical people embed random 3rd party resources. And these 3rd party resources might claim to use only "technically necessary" cookies, but of course that's nonsense; I'm not visiting the 3rd party.

"We have been having extensive discussions around open source strategy. [..] one thing we'd like to do soon is to create a language model with the approximate capability of GPT-3 that can run locally [..]. In general, we think this helps discourage others from releasing similarly-powerful models, and makes it harder for new efforts to get funded." - Sam Altman emails OpenAI board, 2.5 years after GPT-3

Chinese release open models to drive the state of the art, OpenAI and crooked Sam do it to keep you down.

So are computers