It still fails my private physics testing question half the time, where claude 3.5 sonnet and openai o1 (both web version) most of the time passes. So I'd say close to SOTA but not quite. However given deekseek already has the r1 lite preview, and they can achieve comparable performance for much less compute (assuming the API cost of close models roughly represent the inference cost), then it's not unreasonable to believe deepseek may be close to release very good test compute scaling model that is similar to o3 high effort.
Unfortunately, the most likely outcome is they realize this is bad for both, and they should reach a private agreement under the table just like Google Apple no poaching agreement.
Sam Altman said this around a year ago: "i expect ai to be capable of superhuman persuasion well before it is superhuman at general intelligence, which may lead to some very strange outcomes". I am wondering whether AI has any input for this piece that tries to persuade the public to believe the people that are accusing me is not good either.
So you are saying HFT will avoid your market order in this case, while HFT will provide better price when they are the sole counter party in separate liquidity pool? HFT will always maximize profit. To have multiple venues you are just paying HFT as middle man to transfer liquidity from one to another, where you can trade directly with each other if everyone is on one venue, e.g. one centralized limit order book. Transfering liquidity is not HFT's fault, but saying paying for order flow is better for retail is just disinformation. Without evenly discussing the function of HFT, you will get disinformation that demonize HFT as well, and common people won't listen to you later.
I don't really see the need to optimize on 0.03% fee of index ETF such as VOO or VTI. There are much more important things like liquidity, tracking error in whether a ETF is worth buying. Not paying reasonable fee on a service just guarantees some other nefarious ways to get back sooner or later.
It's not. Centralization of liquidity is better for everyone. HFT thrives on fragmentation of liquidity. HFT is not wrong, but fragmentation of liquidity is.
This is intriguing. When I learned that a lot of people do not have inner monologue, I was fascinated by the fact that people can differ on such seemingly fundamental way of being. Maybe those who have it just have a "tee" that pipes into words.
To be honest, I am less worried about the $200 per month price tag per se. I am more worried about the capability of o1 pro mode being only a slight incremental improvement.
Unless you see gold as bullshit as well. I think it should now be consensus that Bitcoin isn't acting as a transaction tool, but as a store of value, just like gold. There are pros and cons when comparing it to gold as a store of value. But the main thing that keeps up the "value" of Bitcoin is just the belief that "it is a store of value" itself. As long as people continue to believe this, it doesn't need to have traditional value. This is the same as gold. Bitcoin in this sense is even more pure, because gold still has some industrial and consumer uses, while Bitcoin has zero. This belief may or may not persist in the foreseeable future, but it is what it is now. Thinking this is "bullshit" just doesn't really matter.
What I find remarkable is that deepseek and qwen are much more open about the model output (not hiding intermediate thinking process), open their weights, and a lot of time, details on how they are trained, and the caveats along the way. And they don't have "Open" in their names.
That's why many ThinkPads have physical covers over their cameras. You don't even need to worry about whether the LEDs are hardwired - relying on any electronic indicator is already a half-baked security measure. If you want real security, just go with a physical solution.
I would expect research along this way to pick up quite a bit if we confirm the pretrain stage is not scaling as previous expected, thus the scale and architecture would be more stable in the near future, especially if the focus shifts to inference time scaling.
I am wondering how much cost is needed for serving at such a latency. Of course for customers, static cost depends on the pricing strategy. But still, the cost really determines how widely this can be adopted. Is it only for those business that really need the latency, or this can be generally deployed.
I thought thid would happen when I see Kalshi launched an almost exact copy of Polymarket but claiming following US laws. This will follow the path of Binance.
I don't think they really "want" you to die; it's just that their priorities do not align with your well-being. Consequently, their actions negatively impact the welfare you could have enjoyed if this power structure prioritized your needs more. Allocating billions to Ukraine and Israel doesn't necessarily directly worsen your situation per se, but it is one of the reasons why the state can only provide $750 USD to families affected by Hurricane Milton if they applications go through.
This seems pretty obvious from Anthropic CEO's recentblog. He has been using words like AI in the hands of "good guys" or "bad guys". You don't need the definition of "good guys" if you don't allow "any" military use. When you start to use good vs bad, it's always something you won't agree with in general.
I think except macros, most of these features are ML family language features as well. Rust stands out because it can implement this in an efficient, zero overhead abstraction way.
Apparently claiming the other side is worse in Gaza issue is not enough. Democratic voters simply refuse to turn out in swing states like Michigan and Wisconsin.
This is such a ridiculous claim. China wants to unify with Taiwan well before US fully allows African American to vote, as guranteed by Voting Rights Act of 1965.
All abstractions are leaky, but some are useful. You can't have a perfect abstraction. Your abstraction is always a compromise of various goals you want to achieve.
I think we need to distinguish between the language e.g. the native language the person uses like English and the concept of language. Your information exchanging binary messages over PCI bus is also part of a language.
I found this FBI whistleblower interview video [1] when I was researching CIA involvement with Xinjiang because I saw a video describing the US strategy about utilizing the Uyghur discontent for US geopolitical gains but could not decide whether this US strategy video was authentic.