I wonder if you could train an LLM with everything up to Einstein. Then see if with thought experiments + mathematics you could arrive at general relativity.
The problem is that the 'genius' of Einstein wasn't just synthesizing existing data,but actively rejecting the axioms of that data. The 1875 corpus overwhelmingly 'proves' absolute time and the luminiferous aether. A model optimizing for the most probable continuation will converge on that consensus.
To get Relativity, the model needs to realize the training data isn't just incomplete, but fundamentally wrong. That requires abductive reasoning (the spark of genius) to jump out of the local minimum. Without that AGI-level spark, a 'pure knowledge pile' will just generate a very eloquent, mathematically rigorous defense of Newtonian physics.
When I hear people complaining about this sort of thing, I want to say, “Just go and invent your own, then.”
But then you get things like Esperanto. Esperanto takes about 1/4 of the time to learn compared to other languages. It’s taught in China and used as primary language in some settings. But, aside from a large number of people learning some Esperanto from Duolingo several years ago, it’s just another language now to have to learn.
This is also about “don’t avoid going to the doctor”. Whether it was an LLM or a friend that “had that and it was nothing”, confirming that with a doctor is the sane approach no?
Which essentially means ignore both the LLM and your rando friend saying "don't worry about it". You shouldn't try to substitute licensed medical evaluation with either.
> [...] shocked that it would suggest and promote false conspiracies, giving support to fantasies that have no evidence, no scientific proof, and just energizes the crazies who already are gullible to misinformation [...]
Incredible the timeline we're in. Companies deliberately adding cognitive poison to their products to... accelerate the dumbing down of society?
Related: All LLMs have struggled helping me generate OpenSCAD models for 3D printing. They can’t seem to reason through XYZ space and 3D math. It’s my current litmus test for LLM performance.
I have had moderate success describing things as geometric primitives. I.e. making a simple phone stand is hard to one shot. Had to do it in steps as make a plane with XYZ dimensions, rotate it X degrees up, calculate the height using trig, create another plane with that height, translate to X position. Etc.
This has been my experience also. Curious if this will improve with just more training data, or if another method will need to be introduced.
Building models in code like this is really cool; it’s great to be able to import, for example, a “gears” library and automatically generate the geometry for complex bevel gear setups. I’ve seen this approach used for more than CAD too; there’s a Python library (GDSFactory) for building photonic ICs in code as well, and I’m sure plenty more.
This workflow is a big motivator for my data notebook project (https://mnty.sh/#serenity), since I would like to build everything for a project in code and have visualizers for each component in one notebook.
One quirk of AI agents is I've moved to `isValid = val > someConstant` over comments because Cursor (I guess Claude by extension) frequently removes and re-writes comments. Or `isValid = checkForValidity(val, someConstant)` if the condition check grows significantly.
This x100. I move it all to a folder called “Unsubscribe” and go through and unsubscribe from everything once in a while.
You can also make it a bit smarter by searching for the header “List-unsubscribe” instead. Less false-positives when someone forwards you an email that contains the word unsubscribe.
You have to be a little careful about this. That works for semi-ethical marketing departments, but for actual spammers it can send a signal that there's a warm body behind the email address, making it far more valuable and more likely to receive even more spam in the future.
I'm starting to think of "vibe coding" as "peer/pair programming". How effective it will be depends on how effective I am as the peer reviewer.
The driver is the AI who is highly capable but has a 5% chance of doing something psychotic lol. Me, the peer, can either review carefully and catch errors or just relax and "vibe" through it all. Results will of course vary based on that relationship.
"Vibe coding" was intended to mean where you don't pay attention to the work your partner creates at all. Where you just lean into their "vibe" and run with it, no matter how bad it actually is. What you describe already has a name. You even mentioned it yourself. Also calling it "vibe coding" would be a bit redundant.
No Vibe Coding is when you dont touch the code yourself and only ask the LLM to make edits for you. You still understand and make sure llm does what you asked, but you dont touch the code yourself.
That wasn't the original intent. It was described as "fully giving in to the vibes, embracing exponentials, and forgetting that the code even exists".
It may have come to mean what you describe to some, but that's more commonly what we call managing a junior engineer. As before, no need for redundancy.
reply