iandanforth

I get the feeling that this guy is an insufferable twat.

Senses are defined by what they sense, not how they are sensed. A person can have their sight partially restored through implants that bypass rod/cone activation. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6934168/

I assumed this was going to be satire. "Let's pre-meeting these TPS reports." Sadly it is not.

The map near my house is so wrong as to be almost unrecognizable.

Totally agree with this article and I'll try to use these words.

Do you think they are using some kind of inaudible-to-humans signal to coordinate this or is the unmodified audio good enough?

Am I misreading this code? It looks like you don't have precomputed table representations and search, instead you scan, embed, and compare on each run?

https://github.com/Dataherald/dataherald/blob/main/services/...

What a terribly argued article and a thinly veiled attempt to undermine the death with dignity movement. There's absolutely nothing I've encountered worse in life than watching someone suffer out their last days. Let's make sure we do our utmost to prevent terminal illness, but when it comes, as it does for us all, every single one of us should have the right to die with dignity.

When was pytype invoked? Runtime? CI/CD? Did it output types to be added to files or did it infer and then enable downstream tasks based on those inferences?

Are there any courses / tutorials / books on the engineering of this system? I'd love to know what kind of choices you make with software when your system needs to run for decades and be remote debuggable at the lowest levels.

Source?

Not at all? Some employers think highly of people with principles.

Lotteries are the only tax that deep red states are in favor of.

They might even get higher accuracies with a dedicated classification layer. By using the existing vocabulary they are spreading the probability mass across a much larger space. If they stuck to N options where N is the total number of functions available to the model I suspect they could get to 100% accuracy.

It's also not clear whether there is sufficient ambiguity in the test data for this to be a generalizable model. The difficulty with "intent recognition" (which they don't mention but is what this problem is called for agents like Siri) is that human generated inputs vary widely and are often badly formed. If they haven't done extensive evaluation with human users and/or they've constrained the functions to be quite distinct then they aren't yet tackling a hard problem, they've just got a complex setting.

It's not. They do train one version with LoRa but also train three variants without.

This is huge.

Along with the paper "Humanoid Locomotion as Next Token Prediction" it seems like the "how to use transformers to get scaling in robotics" problem has been cracked. By using transformers to predict both actions and the next multi-modal inputs (like touch sensation and/or vision) the transformer model learns a useful, general model of the world.

RFM-1 and HLaNTP use totally different models, and data sources but come to the same conclusion, they see scaling behavior similar to those in LLMs. Both are small models compared to LLMs.

This means we're about to see a rapid take off in robot capabilities. 1x, Tesla, and Figure probably have been using this many-to-many transformer backbone for several months now and know just how important it is.

I wouldn't be surprised to see Covariant be acquired quickly to buy their expertise, data, and customer relationships.

This is great, however there were many opportunities to use the word 'nibble' in this post and they were all missed.

https://eyewiki.aao.org/Hallucinogen_Persisting_Perception_D....

"Medicine 3.0" is pretty cringe-worthy though.

Many here will see connections to The Innovator's Dilemma. So I want to ask, are you aware of any business that recognized it had become an impediment to further innovation and "consciously" applied the suggestions from TID? Has anyone done a review of their own business trying to see if they had become part of the problem rather than the best solution to the problem?

We have dozens to hundreds of PRs merged per hour using this flow. Yes the pattern of requiring unchanged main causes a backup. Don't do that.

This is an ad. It doesn't introduce anything new and is probably not a great solution to the posited problem anyway.

Put in-progress code paths behind development flags and merge frequently to main.

That one sentence removes the need for extra tooling and obviates the need for more buzzwordy workflow names like "stacking" or "trunk based" or whatever the trendy thing is at the moment.

These are all very new licenses that deviate from OSI principles, I think it's fair to call them "unusual".

Please note that these results were obtained using a small amount of compute (compared to say a large language model training run) on a limited training set. Nothing in the paper makes me think that this won't scale. I wouldn't be surprised to see a AAA quality version of this within a few months.

Figures 3 and 5 from the paper (https://pubs.aip.org/avs/jvb/article/42/2/023202/3262988/Hig...) have overlap between all the groups. The same measured output could have come from any of the non-cancer, or known-cancer, participants. While the means of these groups look nicely separable, that overlap means there will be significant false positives or false negatives. So I'd label the studyfinds headline of this being 'accurate' to be false.

In the car driving on the mountain road video you can see level-of-detail popping artifacts being reproduced, so I think that's a fair guess.

If you do this for gradient updates you should get a leading indicator.

(again)

For people who enjoyed this post but want to see the other side of the spectrum where self hosted is the norm I'll point to the now classic series of posts on how Stack Overflow runs its infra: https://nickcraver.com/blog/2016/02/17/stack-overflow-the-ar...

If anyone has newer posts like the above, please reply with links as I would love to read them.

I reran a few of the queries that Bard had failed on in Gemini/Ultra and didn't see any improvement. Made the same, or new, logical errors, hallucinated facts, failed to recognize things I was describing etc. It did do better on recognizing an image I uploaded but it went from accurate to nonsensical in the same response.