blueblisters

8 ms·

61.
by blueblisters 1y ago
Nope AGI is not the end goal - https://blog.samaltman.com/the-gentle-singularity > OpenAI is a lot of things now, but before anything else, we are a superintelligence research company. IMO, AGI is already a very nebulous
62.
by blueblisters 1y ago
o3 does better especially if you use the api (not ChatGPT)
63.
by blueblisters 1y ago
This is the best model out there, priced level or lesser than Claude and Gemini They’re not letting the competition breathe
64.
by blueblisters 1y ago
Does this work for simultaneous multiple clients at the same endpoint?
65.
by blueblisters 2y ago
Yeah controlnet-style conditioning doesn't solve for consistent assets, or lighting, framing etc. Maybe its early but it seems hard to get around traditional 3D assets + rendering, at least for serious use-cases. These models do seem l
66.
by blueblisters 2y ago
he had an amazing feel for mass consumer psyche ("taste") and along with a good dose of charisma
67.
by blueblisters 3y ago
SD3.0 looks really good and will probably displace a lot of closed image generation models once it's out. I don't understand why Stability gets so little support from the community. They released the first usable open-source model
68.
by blueblisters 3y ago
How are average headcount costs for an early stage startup $300k / year? That sounds FAANG money
69.
by blueblisters 3y ago
Can you cite a reference for that quote? I can’t find it. He has definitely underestimated language model capabilities but he has been mostly right on the “language models can’t plan” stuff. He has been consistent and correct about his crit
70.
by blueblisters 3y ago
The CEO's job is to enrich the shareholders and by that metric he has done a pretty good job. More qualitatively, being able to change Microsoft's trajectory from boring enterprise tech company to a tech leader with strategic deal
71.
by blueblisters 3y ago
> The point of her work (and mine and Dr. Horvitz’s as well) is to make it clear that putting AI in charge of nuclear weapons and other existential questions is self-defeating Is part of the advocacy convincing nation-states that an AI a
72.
by blueblisters 3y ago
I'm surprised no Chinese automaker has looked at the American market the way Japanese automakers did a few decades ago. Their cost advantage makes it a no-brainer and designing a car to meet American regulations shouldn't be that
73.
by blueblisters 3y ago
I have friends who are Indians living outside. They don’t have a UPI account and find it difficult to get around when they visit India for a few weeks. The simplest solution would be an app that can use a credit card to load to a UPI compat
74.
by blueblisters 3y ago
How are EleutherAI and Carper funded?
75.
by blueblisters 3y ago
Nitpick but how would that be a perverse incentive? That would just be an unintended consequence.
76.
by blueblisters 3y ago
Interesting work! But given the time and complexity of the generation, I think VSD-loss based implementations have higher fidelity. See https://github.com/thu-ml/prolificdreamer or https://github.com/th
77.
by blueblisters 4y ago
> There is something about a critical mass of rule-breakers that sends a system into a negative spiral. Yes this is classic game theory at play. And it’s true for all “low-trust” countries with shoddy rule of law. It doesn’t take decades
78.
by blueblisters 4y ago
What foogazi was perhaps suggesting is that no financial system can work if it doesn't account for fundamental human behavior - we make mistakes and we are greedy if not held accountable "off chain". The blockchain is perfect
79.
by blueblisters 4y ago
This is probably going to have ripple effects in unrelated industries - FTX Foundation, presumably funded by FTX/Alameda/SBF’s personal wealth, has been been the biggest funder of AGI projects and labs this year. They were the big
80.
by blueblisters 4y ago
> My money's on Musk. Same. Worst case scenario for Musk is he gets Tesla to bankroll the takeover with some BS synergies as justification (like he did with Solar City). With fewer engineering resources, I don't expect any maj
81.
by blueblisters 4y ago
More importantly they want to own a platform which won’t make arbitrary rules that upends their entire business model.
82.
by blueblisters 4y ago
In this case the counterfactual is a stronger argument than you saying "they did it only now" with appeal to authority. Unless there's a well reasoned approach to this policy, we are practically stuck with the current standar
83.
by blueblisters 4y ago
This is a very cynical take. Incremental innovation has captured most of the attention in the last decade but there are a lot of smart people in SV working in AI which will lead to real innovation in personal computing and robotics this dec
84.
by blueblisters 4y ago
> To put another one out there, there's no universal and practical way to electronically transfer money from one person to another. The only options are cash, checks, wire transfers and ACH. Wire transfers are expensive and ACH is d
85.
by blueblisters 4y ago
> Everything about banking in the USA was seemingly decades behind my experiences in Northern Europe, so it may take a while for American banks to figure out credit card 2FA… Having worked closely with multiple banks' credit card de
86.
by blueblisters 4y ago
To borrow an analogy from cognitive science, deep learning has pretty much mastered System 1 thinking in independent domains (image, language, audio) and scaling laws suggest we have a clear path to self-supervised multi modal models (video
87.
by blueblisters 4y ago
This is very well articulated and sombre. Even as your neighbouring country, we barely registered anything about the devastation in local news. The recent cricket match between India and Pakistan was an order of magnitude more attention gra
88.
by blueblisters 4y ago
Oh but they have sound which can be annotated much faster/more efficiently. You also potentially have screenplay but the amount of training data is probably too less and sparse. FWIW, I don’t think the AI systems will generate a whole
89.
by blueblisters 4y ago
The flattened image patch of width and height PxP pixels gets multiplied with a learnable matrix of dimension P^2xD where D is the size of the patch embedding. In other words, it’s a linear transformation that reduces the dimensionality of
90.
by blueblisters 4y ago
A majority in a parliamentary system still allows some room for debate and dissent since MPs are, in principle, representatives of their constituency and not of the party. However, in India, it is illegal for an MP to vote against the party

More ›