ankit219

12 ms·

121.
by ankit219 1y ago
Not just Meta, 40 EU companies urged EU to postpone roll out of the ai act by two years due to it's unclear nature. This code of practice is voluntary and goes beyond what is in the act itself. EU published it in a way to say that ther
122.
by ankit219 1y ago
Try one from Kimi 2 as well. I was surprised how good it turned out to be.
123.
by ankit219 1y ago
Seen this happen many times with current agent implementations. With RL (and provided you have enough use case data) you can get to a high accuracy on many of these shortcomings. Most problems arise from the fact that prompting is not the m
124.
by ankit219 1y ago
i would disagree we kind of figured it out early. Early visions for internet were about things like information superhighway (with a centralized approach). What came to pass was the opposite. Its a good thing. There are lessons here in that
125.
by ankit219 1y ago
Think we will see something like this in the near future. But there are two very bold claims: one is hope that RL will lead to generalization in other domains, which based on current evidence seems far fetched. Other is the softwares and sp
126.
by ankit219 1y ago
I can see Gary's point here. He got some stick for this on x, but Gary seems to be right here. One curious thing about this is how both sides have vacated their original positions: scaling people were all about scaling and now they do
127.
by ankit219 1y ago
i see the confusion. Initially the title of the video on BBC youtube channel was: > BREAKING Air India crash: Pilot cut off fuel to engines - no fault with plane | BBC and then they changed it to: > Fuel switches cut off before Air In
128.
by ankit219 1y ago
This is BBC, they have a history of being uncharitable when it comes to reporting on their erstwhile colonies. Jumping to a conclusion about pilot's fault when the recorded dialogues show both had no idea who cut it off (leaving ground
129.
by ankit219 1y ago
Meta offered Pay-or-consent model (nov 23) at 10 euros or so to placate the then GDPR regulators, as the court found contractual necessity as an invalid argument. CJEU stance seems like its valid for meta and they had a long opinion on that
130.
by ankit219 1y ago
Consultation workshops should not be needed. The rule should be clear enough that there is a clear interpretation for everyone. If you need these kind of consultations, you already signal it will be a moving target. Why not just publish cle
131.
by ankit219 1y ago
> Unintentionally installing something from the app store is all good though, because App Store review absolutely ensures that nothing could go wrong, that there's no scam apps, and more than makes up for the web's "orders
132.
by ankit219 1y ago
When a rule is vaguely defined, deliberately so that a regulator can take different interpretations depending on whether they are having any effect and who is doing it, even trivial things become complex. Eg: Meta is asked to withdraw month
133.
by ankit219 1y ago
India has had a history of market manipulation and stock market crashes. One specially high profile one was 1992 after which SEBI got it's powers. While what JS did would likely be allowed (borderline) in many other markets, it would b
134.
by ankit219 1y ago
"Done properly" anything can be science/engineering. Just make a process, give it a KPI on how well it's performing, and you have an engineering. Movie making done properly is where audience is engaged with every act, yo
135.
by ankit219 1y ago
Calling it sloppy work is too charitable. It's one thing for others to give you a benefit of the doubt, it's absolutely crazy that you yourself are doing it. It's clear if the other guy did not speak up, you would not have &q
136.
by ankit219 1y ago
From what it looks like, it's one main LLM (you are sending query to - orchestrator) which calls other LLMs via tool calls. The tools are capable of calling llms too, and can have specific instructions, but mostly just the orchestrator
137.
by ankit219 1y ago
I see this and immediately relived the last two years of the journey. I think some of the mental model that helped me might help the community too. What people expect from finetuning is knowledge addition. You want to keep the styling[1] of
138.
by ankit219 1y ago
They do it seems. o1 and o3 were based on the same base model. o4 is going to be based on a newer (and perhaps smarter) base model.
139.
by ankit219 1y ago
The paper is sloppy. The original point may have credence[1], but the way they went about showing it is borderline irresponsible. The first aspect is how they conflated the number of steps involved with difficulty level. Not even considerin
140.
by ankit219 1y ago
This is very cool. They explained the solution better than I did. If I knew, I would have just linked this :)
141.
by ankit219 1y ago
Thats how a lot of application layer startups are going to make money. There is a bunch of high quality usage data. Either you monetize it yourself (cursor), get acquired (windsurf) or provide that data to others at a fee (lmsys, mercor). T
142.
by ankit219 1y ago
I have a different pov on retrieval. It's a hard problem to solve in a generalizable format with embeddings. I believe this can be solved at a model level where its used to fix an issue. With the model providers (oai, anthropic) going
143.
by ankit219 1y ago
I simplified what i wrote. There is an off accelerator memory where the embeddings are stored and queried at inference time, i did not want to get into details. That is how you reduce the in memory RAM. There are definitely more things goin
144.
by ankit219 1y ago
You can read this for a comprehensive deep dive. https://arxiv.org/pdf/2502.01637 At a very high level, instead of having embeddings at the input layers, this method keeps the embeddings at the layer level. That is eve
145.
by ankit219 1y ago
This is very cool. One question I have in these orchestration based multi agent systems is the out of domain generalization. Biotech and Pharma is one domain where not all the latest research is out there in public domain (hence big labs ha
146.
by ankit219 1y ago
This is a very millennial style of thinking (myself included). It feels like people can't just have a hobby, they have to be great at it. The sense of greatness, the sense of accomplishment is not merely doing a thing, but getting to a
147.
by ankit219 1y ago
Familiar scenario. One thing that would help is to use AI as someone you are having a conversation with. (most AIs need to be explicitly told before you start this). You tell them not to agree with you, to ask more questions instead of prov
148.
by ankit219 1y ago
Cynicism aside, Apple also has a very critical trust problem to solve here. Many non tech users in Apple ecosystem are used to a specific flow: See an interesting game, download it, when the game asks to pay, link your apple wallet, and the
149.
by ankit219 1y ago
Not to these companies who want it. To Google yes. Not exactly because they have to, but also because they built things people want.
150.
by ankit219 1y ago
Woudl the search index contain only raw data about the websites? Or would some sort of ranking be there? If it's teh latter, its a neat way to ask a company to sell their users data to a third party because any kind of ranking comes vi

More ›