I’m starting to see huge solo projects with 100 commits per day turning up on GitHub.
Last year these were unusable slop.
Now they’re “getting there”. Not quite as good as hand crafted code written by humans, but usable.
30,570 karma · joined December 4, 2018
Last year these were unusable slop.
Now they’re “getting there”. Not quite as good as hand crafted code written by humans, but usable.
today: https://www.anthropic.com/news/redeploying-fable-5
35 days ago: https://www.anthropic.com/news/claude-opus-4-8
70 days ago: https://openai.com/index/introducing-gpt-5-5/ <-- first model I've found useful
77 days ago: https://www.anthropic.com/news/claude-opus-4-7
119 days ago: https://openai.com/index/introducing-gpt-5-4/
182 days ago: The start of the year
This may as well have been written in the stone ages, when we were banging AI rocks together.
I just did a ~6 month project in ~2 weeks using a frontier model.
I wouldn't even have attempted this kind work a year ago, with or without the AIs available at the time!
When “not cartoonishly corrupt” becomes hard to believe due to firmly established character and endless precedent, then it almost doesn’t matter if everything is on the level in this one matter.
The trust has been lost, undermining the ability to govern.
The fact that he’s still in power with firm support from everyone else that matters is all you need to know about how rotten to the core things are in your country.
America is no longer the greatest, or the best.
Except at grift…
Sure… but which ones? How can you know ahead of time?
I just did a “simple” upgrade project where both me and the AI kept tripping over dead code, subtle typos, and difficult-to-trace live versus dead code.
Many times I used “Medium” thinking I got bitten, but not every time, and I couldn’t predict when.
So “Extra high” it was, for the entire project.
Far fewer nasty surprises!
It's the same logic as keeping RAW digital photos. Lightroom is already gaining "find the keepers" AI features. Sooner or later it'll be possible to feed a bunch of burst shots into an AI that just weren't worth the trouble to manually sort through. The AI can do the drudge-work of digging up any gems in the rough...
Saying that “less diluted works better” is saying the diametric opposite of what Homeopathic “theory” does!
Other nations are falling behind and will be at a real disadvantage soon.
That meat follows an ill-defined pattern encoded in fewer bits than the source code of PyTorch and its pretraining phase used a tiny fraction of the available data.
You’re a poor imitation of an LLM.
I mean… you’re fluent in, what, at most five or six languages? Can program in maybe another dozen if we’re being generous about your capabilities?
Pfft… who would trust anything to meat brains!? They’re famously prone to hallucinations!
I was just saying to colleagues that I haven't felt the need to go past an 8 core machine until this month, when I started running parallel GPT 5.5 agents on a decent sized codebase (over 4 MB of code). There were times I could barely move my mouse cursor!
One coder at $dayjob had a “thing” where they wrapped every single block of code in a try-catch block. Thousands upon thousands of instances, enough to bloat the code volume 2x and slow down compile times.
I got an AI to write a cleanup tool that internally used a “compiler toolkit” (the Roslyn SDK) to mechanically and precisely parse the source and safely perform the substitution.
It ran for fifteen minutes, made about ten thousand edits, and halved the size the codebase with zero errors or side effects.
Not my problem!
Over here we depose of our leaders when their corruption comes to light.
The "Plus" and higher tiers are advertised as "The latest models" but if that's not true any more then Open AI is opening themselves to investigations from organisations like: https://www.accc.gov.au/
Promising "X" for $Y to everyone and then delivering "X" only to the "chosen few" at the detriment of others opens you up to lawsuits because then your product advertisement is a lie.
For a comparison, imagine a telco selling "100 Mbps for $20/mo" and then throttling the connection for customers that aren't currently favoured by the Trump administration!
Trump and his cronies.
Nobody else in the US Gov is so openly corrupt and has the unitary executive power to ban the products of privately held companies for sale without there being a public record (i.e.: an act of congress).
I recently used GPT 5.5 "extra high" running nearly non-stop for about a week to upgrade a legacy ASP.NET Web Forms app to ASP.NET Core on .NET 10.
This was considered "too hard" (too expensive) for human developers because it is a wholesale rewrite of every web page template. Not to mention that the dependency injection mechanism is totally different, async is more pervasive, etc.
Worse still, the old app was split into a bunch of components with a variety of web API protocols in between them, had stupidly complex Oracle stored procedures, and a whole series of hidden land mines in the codebase. It was an undocumented, unmaintainable mess full of dead and spaghetti code.
I would have estimated 6 months minimum for a human developer to uplift it, but 12 months is more realistic.
Doing the same in a week feels like tapping into some sort of forbidden black magic. It feels like I can't admit this to anyone, lest they think I dabble in the dark arts.
You'd be shocked how often I see the meat-space equivalent of vibe coding!
"I trust the developers."
"You really shouldn't!"
The thing to realise is that there is no fundamental difference between outsourcing a development task to other human developers versus outsourcing[1] it to LLMs.
Either way, total and complete understanding is being sacrificed in the name of productivity and scalability.
It's just there's one extra layer of work assignment now, with ICs handing off tasks to agents.
What this has revealed to ICs is the BIG issue that has plagued all software development for decades, especially since outsourcing became so popular: Oversight is critical, and more importantly: authority can be delegated, but responsibility cannot.
LLM output is fine, as long as you review everything it does.
This is the same as any competent dev team manager reviewing PRs for quality, paying attention to critical matters such as security, adherence to high level design and low-level style standards, etc.
Some do.
Many never did.
[1] This doesn't have to be a contract with an overseas provider, by "outsourcing" I mean any variant of not-your-own-hands-on-keyboard. Any scenario where a customer or manager assigns tasks to developers other than themselves.
Sure, but landing is not necessary for doing much more science than we are right now!
An orbiter dedicated to Io would dramatically increase our knowledge of this unique and special moon.
Similarly, you obviously couldn't use exact replicas of the same design for the space probes, you'd need variants to deal with the greater distance from the Sun.
You would need close/medium/far probe designs with small/medium/large solar panels, radio antennas, and camera optics to compensate for the lower light levels.
For each design variant you could make 10-20 exact clones, and the variants would be very similar to each other.
We do this now!
Just not for science.
Spy satellites tend to be serially produced, as are GPS satellites, and of course Starlink satellites, which are made in their thousands.
So instead of designing totally bespoke multi-billion-dollar probes, build dozens of hundred-million dollar probes that leverage efficiencies of scale just like any other industrial product.
Send them to every outer planet, every moon, and every large asteroid like Ceres!
That bug caused untold grief for multi-user session hosts / terminal servers / Citrix XenApp for a better part of a decade. Way before slopware!
People forget that software written by people is… even worse.
It’s like the self driving car debate: sure, the robot taxis will kill people on occasion, but people kill people regularly.
It's a shortcut useful only if you need to scribble on paper and your wrists hurt from writing too much, but it obscures the underlying physics.
The programming equivalent is putting abbreviations in identifiers where, sure, it's fewer characters, but then the reader needs to a track a mental lookup table to translate back to the intended meaning.
Pushing things like this too far results in meaningful aspects of the equations getting squeezed out entirely. For example, the generality of GA means that you have to (correctly) track negative signs and multiplications by pseudoscalars such that your formulas work in all dimensions. In traditional vector algebra it's all too tempting to eliminate certain products because in "your chosen dimension" they multiply to 1 or -1 or whatever and just... disappear due to traditional algebraic simplification conventions. But then if you need to work in 4D SR or curved spaces, you can't, because you threw away something essential while "optimising for characters on a page".
You have then "start over", typically reaching for a partial and incomplete subset of GA, reinventing that wheel over and over.
Hence the push for unification onto GA, to break this cycle.
The main advantage is the elimination of gimbal lock which allows smooth interpolations of arbitrary rotations and translations. This dramatically simplifies certain codes relating to animation and robotics. In mathematics, it simplifies differentiation and integration over curved surfaces in 3D and higher dimensions.
Most developers working in those fields already use one of the many creatively named “impostors” of GA subsets in isolation, such as the quarternions.
The point I’m trying to make is that there are necessarily complexities inherent in all areas of study, and there are incidental complexities because of historical reasons, “culture” within certain fields, or juniors putting out their fields’ equivalent of spaghetti code.
Geometric Algebra sweeps away a lot of the rather messy parts of now century-old physics, but the work of doing that substitution is decidedly non-trivial and thankless, so other than Hestenes, nobody seems to be pushing for it.
It’s like the 2pi versus tau fad on the internet.
Mathematicians argue that they’re “the same”, so it doesn’t matter, and ramble on about their equivalent of “learn the Latin to be smart like me”.
No. It’s stupid. It was an error. Tau is the correct circle constant and eliminates magic constants that don’t belong from literally hundreds of famous formulas!
I and many others simply failed to understand radians until I learnt to treat 2pi as a single ligature instead of “two of something”.
Certain kinds of perfect correctness are like pure and shining crystallised bits of refined knowledge created by the greatest wizards. "Parse, don't validate" or "Make invalid states unrepresentable." ought to be familiar to the better programmers here, the ones with decades of experience built on iterative, collaborative foundations with real consequences for error.
Theoretical physics doesn't have those same consequences, because there is no real punishment for their equivalent of "spaghetti code". Perversely, there's cachet to be gained for gaining understanding of its unnecessarily esoteric knowledge, much like how biologists and lawyers spend half a decade or more studying... Latin.[2]
Introducing Geometric Algebra to physics is like that wizard coder who sweeps away reams of spaghetti code and replaces it all with a call to a single standard library function. It's that "cheff's kiss" of cleanup. Meanwhile the juniors are screaming about how the senior "deleted all their hard work!"
Meanwhile, I never understood where Pauli and Dirac matrices came from! It's like they were pulled from fat air.
You've seen this in code, I bet. Some junior worked really hard on solving a problem and wrote a solid screen-filling wall of "a && b || c || !d && e && (f || g)..." continuing up to "ba, bc, bd", etc.. as they ran out single letters until they're well into the alphabet in double-character symbols.[3]
That's what those matrices are. Someone's hacky attempt at "making things work".
The problem is that we gave those people Nobel prizes and told everyone they're geniuses.
They are, but they were like that brilliant junior. Brilliant.. but junior.
Geometric Algebra sweeps all of that into one beautiful, consistent, crystal clear abstraction that is widely applicable. The magic matrix constants vanish. Bugs in 100-year-old textbook formulas suddenly come to light. Dozens of formulas, one set for each of the 1D, 2D, 3D, and 4D cases collapse into a single formula valid for any number of dimensions.
It's like watching someone struggle with "catching every possible instance of JavaScript injection".
No son, no. Just no. Stop enumerating badness. Stop. Just stop. Escape everything at the boundary instead, enforced by the type system. You'll thank me later.
I know it might be obvious to you, and you always use properly parameterised SQL queries or whatever. This is not the norm everywhere! I still get arguments, long drawn out arguments from people convinced that this is unnecessary and just one more search & replace is all they need to be safe from the bad hackers.
Physicists (and mathematicians) are still making that argument against GA.
"It's isomorphic!"
"That isn't the point!"
[1] You can't convince someone to climb Everest if they struggled to hike up to the top of one of its foothills.
[2] Let me be crystal clear: They're spending their precious time on this Earth learning a dead language instead of learning about the law or bugs. No amount of arguments will sway me. The bugs don't care what you call them. Criminals are guilty or innocent whether or not you speak funny in court. You've just made a simple thing harder for no good reason, that is all. Please stop.
[3] Yes, I've seen this. Twice, from two different people whom have never met. Aliens are amongst us.
It makes hybrids like this very dangerous for anything even remotely security-adjacent, such as roles, tokens, etc.
This kind of thing caused the CVE-2009-2408 and CVE-2009-2510 "Null Truncation in X.509 Common Name Vulnerability."
a) There must be network packet buffers on each receiver, which is a per-server "queue".
b) The optimal performance can only be achieved by pipelining requests sent to the servers, otherwise there is "dead time" caused by the network latency.
Perfect queuing to workers that act as a "slot" can only be implemented in-process on a single machine, and even there having small queues or buffers per core can improve throughput!