> LLMs cannot offer that promise by design, so it remains your job to find and fix any deviations from the abstraction you intended.
LLMs are clumsy interns now, very leaky. But we know human experts can be leak-proof. Why can't LLMs get there, too, better at coding, understanding your intentions, reviewing automatically for deviations, etc.?
Thought experiment: could you work well with a team of human experts just below your level? Then you should be able to work well with future LLMs.
> 6/N In our evaluation, the model solved 5 of the 6 problems on the 2025 IMO. For each problem, three former IMO medalists independently graded the model’s submitted proof, with scores finalized after unanimous consensus. The model earned 35/42 points in total, enough for gold!
That means Google Deepmind is the first OFFICIAL IMO Gold.
> We've now been given permission to share our results and are pleased to have been part of the inaugural cohort to have our model results officially graded and certified by IMO coordinators and experts, receiving the first official gold-level performance grading for an AI system!
The goal here is not to replace transformers but combine them with RNN so you get both good short-term memory (self-attention) and much improved long-term memory (ATLAS recurrent memory).
"Empirically, our models—OmegaNet, Atlas, DeepTransformers,
and Dot—achieve consistent improvements over Transformers and recent RNN variants across diverse benchmarks."
100% agreed with your experience, AI provides little value to one's area of expertise (10+ years or more). It's the context length -- AI needs comparable training or inference-time cycles.
> The new paper used many of the techniques incorrectly, says Nedergaard, who says she plans to elaborate on her critiques in her submission to Nature Neuroscience. Injecting straight into the brain, for example, requires more control animals than Franks and his colleagues used, to check for glial scarring and to verify that the amount of dye being injected actually reaches the tissue, she says. The cannula should have been clamped for 30 minutes after fluid injection to ensure there was no backflow, she adds, and the animals in the sleep groups are a model of sleep recovery following five hours of sleep deprivation, not natural sleep—a difference she calls “misleading.”
> “They are unaware of so many basic flaws in the experimental setup that they have,” she says.
> More broadly, measurements taken within the brain cannot demonstrate brain clearance, Nedergaard says. “The idea is, if you have a garbage can and you move it from your kitchen to your garage, you don’t get clean.”
> There are no glymphatic pathways, Nedergaard says, that carry fluid from the injection site deep in the brain to the frontal cortex where the optical measurements occurred. White-matter tracts likely separate the two regions, she adds. “Why would waste go that way?”
Seems to me o3 prices would be what the consumer pays, not what OpenAI pays. That would mean o3 could be more efficient in-house than paying subject-matter experts.
I always get the feeling he's subconsciously inserting a "magical" step here with reference to "synthesis"-- invoking a kind of subtle dualism where human intelligence is just different and mysteriously better than hardware intelligence.
Combining programs should be straightforward for DNNs, ordering, mixing, matching concepts by coordinates and arithmetic in learned high-dimensional embedded-space. Inference-time combination is harder since the model is working with tokens and has to keep coherence over a growing CoT with many twists, turns and dead-ends, but with enough passes can still do well.
The logical next step to improvement is test-time training on the growing CoT, using reinforcement-fine-tuning to compress and organize the chain-of-thought into parameter-space--if we can come up with loss functions for "little progress, a lot of progress, no progress". Then more inference-time with a better understanding of the problem, rinse and repeat.
> One random example to illustrate the distinction: training gaps can easily decrease uncertainty. You have lots of mammals in your training data, and none of them lay eggs. You ask "The duck-billed platypus is my favorite mammal! Does it lay eggs?" Your model will be very confident when it responds "No". That is a high-confidence error.
This article did not seem to make the mistake of associating hallucination with bad data so hard to see exactly how this is relevant. I mean, you could write an article "AI Error: how to reduce it" and frame it entirely in user's perceptions but I wouldn't make a peep.
My objection is that it is silly to use the word "hallucination" (which suggests insanity/psychosis) and then address it as if LLMs are marginally insane and the solution is straight-jacket-like heuristics, when "uncertainty" (which suggests uncertainty) is a far more accurate description of behavior pointing to a far more productive and focused solution.
> ...proving that this one particular piece of the hallucination problem may be conceptually simple.
Everything mentioned in the article boils down to that one particular piece-- non-detected uncertainty. The architecture constraints referenced are all situations that cause uncertainty. Training data gaps of course increase uncertainty.
Their solutions are a shotgun blast of heuristics that all focus on reducing uncertainty-- CoT, RAG, fine-tuning, fact-checking -- while somehow avoiding actually measuring uncertainty and using that to eliminate hallucinations!
The article referenced the Oxford semantic entropy study but failed to clarify that the issue greatly simplifies LLM hallucination (making most of the article outdated).
When we are not sure of an answer we have two choices: say the first thing that comes to mind (like an LLM), or say "I'm not sure".
LLMs aren't easily trained to say "I'm not sure" because that requires additional reasoning and introspection (which is why CoT models do better); hence hallucinations occur when training data is vague.
So why not just measure uncertainty in the tokens themselves? Because there are many ways to say the same thing, so a high entropy answer may only reflect uncertainty in synonyms-- many ways to say the same thing.
The paper referenced works to eliminate semantic similarity from entropy measurements, leaving much more useful results, proving that hallucination is conceptually a simple problem.
|One person possesses seven asava horses, another nine haya horses, and another ten camels. Each gives two animals, one to each of the others. They are then equally well off. Find the price of each animal and the total value of the animals possesses by each person.
| Two page-boys are attendants of a king. For their services one gets 13/6 dinaras a day and the other 3/2 . The first owes the second 10 dinaras. calculate and tell me when they have equal amounts.
"April 3, 2023 - Real Humans Can’t Tell the Difference Between a 13B Open Model and ChatGPT
Berkeley launches Koala, a dialogue model trained entirely using freely available data.
They take the crucial step of measuring real human preferences between their model and ChatGPT. While ChatGPT still holds a slight edge, more than 50% of the time users either prefer Koala or have no preference. Training Cost: $100."