jpcompartir
I suppose when people do re-write the textbooks they'll say "this is confirmed now" not "if this conjecture is true...", but usually re-writing the textbooks would imply that things have been shown to be false?
May have misunderstood. Thank you for the post though, it was very interesting to someone who doesn't know much about the topic.
I hope he still gets to do some educative stuff on the side too
Recently that immaculately polished feel is harder to find. It coincides with the daily releases of CC, Desktop App, unknown/undocumented changes to the various harnesses used in CC/Cowork. I find it an unwelcome shift.
I still think they're the best option on the market, but the delta isn't as high as it was. Sometimes slowing down is the way to move faster.
But this week I've lost count of the times I've had to say something along the lines of: "Can you check our plan/instructions, I'm pretty sure I said we need to do [this thing] but you've done [that thing]..."
And get hit with a "You're absolutely right...", which virtually never happened for me. I think maybe once since Opus 4-6.
So the bottleneck was compute. Which is compatible with 'data or compute'. But to accept your point, at the time the algorothmic advances were useful/did unlock/remove the bottleneck.
A wider point is that eventually (once compute and data are scaled enough) the algorithms are all learning the same representations: https://arxiv.org/pdf/2405.07987
And of course the canon: https://nonint.com/2023/06/10/the-it-in-ai-models-is-the-dat... http://www.incompleteideas.net/IncIdeas/BitterLesson.html
Scaling compute & data > algorithmic cleverness
The bottleneck in AI/ML/DL is always data (volume & quality) or compute.
Does/can Autoresearch help improve large-scale datasets? Is it more compute efficien than humans?
Always rooting for Hugging Face
As others have said - 'reading 10 files' is useless information - we want to be able to see at a glance where it is and what it's doing, so that we can re-direct if necessary.
With the release of Cowork, couldn't Claude Code double down on needs of engineers?
This won't change my decision, but it is still impeccable timing
Everything in plan mode first + AskUserQuestionTool, review all plans, get it to write its own CLAUDE.md for coding standards and edit where necessary and away you go.
Seems noticeably better than 4.5 at keeping the codebase slim. Obviously it still needs to be kept an eye on, but it's a step up from 4.5.
It's a very powerful way to work on all kinds of things. V. interested to try co-work when it drops to Plus subscribers.
Which isn't particularly amazing, as # of tokens generated is basically a synonym in this case for computation.
We spend more computation, we tend towards better answers.
https://cdn.openai.com/pdf/d04913be-3f6f-4d2b-b283-ff432ef4a...
They attribute these 'compression artefacts' to pre-training, they also reference the original snowballing paper: How Language Model Hallucinations Can Snowball: https://arxiv.org/pdf/2305.13534
They further state that reasoning is no panacea. W hilst you did say: "the models mitigate more and more"
You were replying to my comment which said:
"'Bad' generations early in the output sequence are somewhat mitigatable by injecting self-reflection tokens like 'wait', or with more sophisticated test-time compute techniques."
So our statements there are logically compatible, i.e. you didn't make a statement that contradicts what I said.
"Our error analysis is general yet has specific implications for hallucination. It applies broadly, including to reasoning and search-and-retrieval language models, and the analysis does not rely on properties of next-word prediction or Transformer-based neural networks."
"Search (and reasoning) are not panaceas. A number of studies have shown how language models augmented with search or Retrieval-Augmented Generation (RAG) reduce hallucinations (Lewis et al., 2020; Shuster et al., 2021; Nakano et al., 2021; Zhang and Zhang, 2025). However, Observation 1 holds for arbitrary language models, including those with RAG. In particular, the binary grading system itself still rewards guessing whenever search fails to yield a confident answer. Moreover, search may not help with miscalculations such as in the letter-counting example, or other intrinsic hallucinations"
As of today, 'bad' generations early in the sequence still do tend towards responses that are distant to the ideal response. This is testable/verifiable by pre-filling responses, which I'd advise you to experiment with for yourself.
'Bad' generations early in the output sequence are somewhat mitigatable by injecting self-reflection tokens like 'wait', or with more sophisticated test-time compute techniques. However, those remedies can simultaneously turn 'good' generations into bad, they are post-hoc heuristics which treat symptoms not causes.
In general, as the models become larger they are able to compress more of their training data. So yes, using the terminology of the commenter I was responding to, larger models should tend to have fewer 'compression artefacts' than smaller models.
"Backpropagation, often referred to as “backward propagation of errors,” is the cornerstone of training deep neural networks. It is a supervised learning algorithm that optimizes the weights and biases of a neural network to minimize the error between predicted and actual outputs.."
https://chizkidd.github.io/2025/05/30/backpropagation/
backpropagation is a supervised machine learning algorithm, pardon?
And if we increase N enough we will be able to find these 'good measurements' and 'statistically significant differences' everywhere.
Worse still if we did not agree in advance what hypotheses we were testing, and go looking back through historical data to find 'statistically significant' correlations.