Yeah and anthropic can't train Claude better than anyone else either so no moat there either. Knowledge is not scarce it just takes work to uncover once uncovered it is not useful if not published. If published it can't be owned.
In general llms are weak with spatial reasoning. This seems to be an unsolved problem. Probably because human language is generally imprecise spatially and humans think about spatial problems in visual terms. I wonder if having an llm make a 3d design in a format an image model could check would result in a better outcome?
Having used them heavily for 3D model understanding since February I can say it's nuanced.
Opus 4.6 and 4.7 were bad, but GPT 5.2 and above were very usable. Opus 4.8 was usable, but the GPT 5.x series was better.
Fable is great.
Opus 5.0 was interesting. It could solve some problems that Sol 5.x couldn't solve (applying a G2 curve on a 3 way corner where one face was a Bezier curve) but you had to be super prescriptive ("only answer the question"/"only do what I tell you and stop when done") or it would go on a hugely involved validation journey that didn't really achieve a lot.
Opus 5.5 is better than that was in that respect.
Sol 6.x is great, and my daily driver for this (I use Opus for coding though)
Astra can solve problems that Sol can't but for some reason on easy stuff makes uglier solutions.
For all models it's very interactive though - we aren't at the "agentic design" phase for most things yet.
Is it that LLMs are weak with spatial reasoning (and memory) or is it that we are unusually good at it?
When I need to use a program I seldomly use I'm far more likely to remember where I need to click to open it than the word I need to search for to open it.
Yes I like to think of humans with built-in accelerators for certain tasks -- our visual and spatial reasoning is off the charts presumably because it's a life or death skill!
I made a building and had astra fill out the interior of the bathroom with toilets. It put 6 of them in two rows back to back with no way to reach the second row. Other than climbing over the stalls of the first row I suppose.
I think that this is an unsolved problem in the same way that mangled fingers in image generation was an unsolved problem.
Through at least Opus 4, LLMs were practically useless for authoring any sort of coherent procedural closed-curve geometry (I know this with strong confidence because of the little animated guys at https://letterspractice.com).
Opus 5.5 can bang it all out. Possibly a deliberate RL sort of thing or maybe another surprise emergent capability.
There's an interesting comparison in the creative/literary end of llm output too, they're in my experience, dreadful with anatomy. Like, it knows humans have hands, heads, etc, but often times a seemingly limited concept of how anything is connected, or degrees of freedom. (e.g., Why yes, certainly there are many examples of humans rotating their torsos 180º at the hip, seems perfectly cromulent)
I honestly don't know if an image model would help, or if it might analyze the output and go "13 fingers? ship it!" anyway.
LLM output should be credited to whoever prompted it. I don't get this hysteria. Llms are tools like compilers or chain saws. We don't say a cabinet made from milled wood is some how less authentic than a cabinet made with flint tools. We don't deny people who write compiled programs in a high level language without understanding every branch of the AST and every transistor setting authorship of their programs. Just because something is somewhat non deterministic and draws from the wealth of previously curated human knowledge does not make it an independent agent and not a tool. The people who find AI generated work 'sad' are implicitly sad about the democratization of knowledge and the lowering of barriers to entry to producing things. The thing produced is still a human product. Rather than saying it is an ai creation we can say it was author prompted and compiled of the combination of the author's intent and the store of human knowledge and reasoning trained into an LLM. What is there to be sad about a new tool existing to make things?
I used to think this, but I can ask an LLM to design me a CPU at this point while knowing absolutely nothing about how it works. Literally nothing. This is not the same as writing a program and handing it to a compiler. This is more like asking youtube to give you a video, then pretending you are Mr. Beast.
Thanks for making my point for me, although probably not in the way you intended.
The credit for the Doordash meal should be given to the person placing the order. If it weren't for their amazing prompting^Wordering skills, which can make food materialize out of thin air, we would not be eating today.
Yes an usually it is easy to tell one shot ai attempts vs careful or interatively regimes designs. Just because ai was used to execute and help generate the design in detail does not mean the human in the loop is not essential to the product.
"LLM output should be credited to whoever prompted it" promptly falls apart when you convert it to its pre-LLM equivalent "worker output should be credited to the manager who supervised it." These LLM bros are like bad managers trying to pretend the creativity, reasoning, and effort of their subordinates is actually their own just because they had the idea.
My understanding is no llm doe anything without prompting. Ye agent can self prompt and engage in loops but eventually that run down and usually it tay pretty close to the original prompt or soul document. Just like with ai math there is zero agency here. Nothing is happening without human direction. If you give agenda access to the Internet and a goal like pen test government websites they will if not restricted by guardrails. Nothing is suprising here. There is a tool working under human direction. If a person sets up a trap in the woods and it maims someone else we don't say the trap is responsible.
Also the policy makers that made infrastructure and laws susceptible to automated manipulation which decidedly is not the case today. For example LLMs are not able to manipulate the power grid because they are not exposed to the control interfaces..no matter how intelligent llms get their physical levers do not extend unless people decide to let them. The ai threat is real ridiculous.. yes you may ee ome uplift in bad actor. Like for example f someone wants to build a tank dozer in their garage and is not experienced welding ai might give some useful information as a book would in the past. However just like a book ai is bot going to build and operate it itself unless someone design and effective android robot. Once LLMs or other ai becomes an embodies agent then there may be a real threat if agency. Last time I checked ai can't drive a car but it can tell you to walk your car to a car wash.
It is an elaborate business ploy to create a regulatory framework for "safe-ai" that shields these companies from liability. This way, they can sell "safe-ai" to enterprises and if shit-hits-the-fan at the enterprise, sorry, this is certified "safe-ai" so, your bad. Shift blame to a regulatory body. From an enterprise buyer's perspective they can say, hey, I bought "safe-ai" and so dont fire me when it "rm -rfs" the production database.
Still, beats me why they are painting their product in a negative light, and scaring their own enterprise customers. After this sort of marketing, any enterprise buyer would be scared to go anywhere near it.
Depends on where they are from. Iranians from the north, the caspian coast can look very Nordic. I would say the average Iranian looks about as white as the average Italian or French man.
Honestly I am not impressed with Tao treating "ai" as a threat to mathematicians. He might as well say math itself, specifically linear algebra and gradient descent are a threat to mathematicians. Maybe think of something original rather than pander the hype of day?
Can we get a bullshit kill switch where if your company consumes so many dollars for so long it is dissolved. Bullshit is a far bigger threat than imaginary rogue AI.
