Vishal, your opinion here borders upon disgusting.
Be glad you don't understand true depression. Be glad that you know nothing of these people's minds, and shut your dumb fucking mouth.
There's no way to respond politely to somebody blithely writing the ignorant horseshit you wrote. You're a fucking idiot, and you deserve to be treated with the utter contempt and lack of respect that you have for others.
Imo the title here buries the lede: we've known that trees form tension wood in response to external stress. This study claims to find that trees will form it in response only to their own curvature, independent of external factors even including gravity. They compare it to proprioception. That's neat!
Yes, catastrophic forgetting is absolutely one of the problems that needs to be solved to enable something like this.
My broader point is just that there's nothing inherent to the structure of LLMs that stops them from updating their weights and continuously learning from environmental feedback in the way humans do, and there's already solid templates for how they could push even further in that direction.
But as an assessment of the current state, I agree with you, LLMs lag humans severely in ability to self-update.
I mean sure, but BERT can be and often is used as a next token predictor / generator.
I could have used any series of NLP examples, the point is this CoT behavior only emerges when you get to a certain scale (and training style, presumably).
That is backwards. It is the responsibility of a researcher to do a thorough literature review and conscientiously avoid plagiarism or claiming false novelty.
Carney is one of them few (if there are any) leaders of Western countries who is actually genuinely popular and has a very high approval so he does have a lot of leeway to do what he thinks is best.
This is where I'm at as well. I've been using AI increasingly more and more at work and it has been genuinely useful. I want a Siri-like assistant that is actually useful. Muse seems like it's that tool. Except, it's Meta.
That's not quite true though. The fact that most models are in practice non-deterministic has been a huge point of contention.
And there's nothing inherently stopping labs from continuously fine-tuning the weights after every new invocation. It's just a difficult (not to mention expensive) software and ML problem.
Excellent information. Unfortunately, I do not have the knowledge to address the issues identified with my home network although I do suspect unauthorized remote or virtual access.
kinda crazy that just switching drives can make an unreadable disk readable
makes me wonder how many old floppies got written off as dead when it was really just an alignment mismatch
Full-stack and AI engineer, 3 years shipping to production at an EV-charging platform used by 50k+ drivers - LLM features, tool-using agents, Node/TypeScript services, Postgres, and the operational side of keeping them correct. IIT Kharagpur, 2023.
Recent: a WhatsApp AI receptionist running live - Cloud API webhook, an agent that answers from the business's own content, captures leads and books appointments, with a clean human handover. https://brightsmile-whatsapp-agent.vercel.app
The work I am best at is the part after the demo: idempotency so a redelivered webhook does not double-book, durable records written before any external write, and evals on an agent's decisions rather than eyeballing a few transcripts.
Open to full-time roles and to contract work (part-time, IST evenings and weekends).
I think that's relatively emergent too though! BERT never really did that (at least to my recollection), presumably because its training was never sufficient for it to develop corrective reasoning in a chain of thought.
The Target marketing team famously could predict when a woman is pregnant but it was too creepy to, uh, Target them with only baby goods in dynamic marketing materials so they would mix in a few masculine things like lawn mowers into the catalog. No idea if that is what Meta is doing here but it's one explanation.
I watched the video and they're barely really trying to hide it. It reads some emails and immediately looks for stuff to advertise to the humans. Imagine how biased it will be once enshitification begins.
Very cool. I was expecting to learn about the ins and outs of ink, printheads etc but instead learned "How to build a paper that acts like a printer" which turned out much more entertaining.
Here's a wake up call for everyone sending all of their ip to openai and anthropic. Especially in verticals they intend to dominate. Lol at all the biotech companies all in on Claude and paying millions in fdes creating huge lapses in security as they go.
As an MLE I feel these takes are too reductionist.
You could say the (nearly) same thing about search. And content recommendation. And clustering. And topic modeling. And outlier detection. And spam filtering. And image diffusion. And dimension reduction. And...
There's a lot in common between these things, but there's also a lot cool and different!
For transformers in particular, it's pretty cool that you get some WILD emergent properties simply from scaling up.
So yes, it's just a next token predictor, but I'm just a bundle of nerves and meat. I don't get a lot out of those descriptions.
Pretty happy with the experience of using it so far. Asked it to do something to blow me away and it made a podcast about the features it made while onboarding itself to my opensource project called DashClaw.
Then I had it make a marketing video as a test, pretty neat. Obviously all of this was doable but quite impressed with the model underneath since it's all free, for now.
Be glad you don't understand true depression. Be glad that you know nothing of these people's minds, and shut your dumb fucking mouth.
There's no way to respond politely to somebody blithely writing the ignorant horseshit you wrote. You're a fucking idiot, and you deserve to be treated with the utter contempt and lack of respect that you have for others.
I hope you get hit by a bus and live.