22,069 karma · joined January 10, 2013
It sure sounds like a PyTorch tutorial, but I believe it's yet another "AI training made slightly easier for you" start-up. But all of them seem to solve the easy problem of managing data and compute, while very few tackle the hard problem of generating good training data.
Public companies will happily can the entire management team if they cause as little as a 5% dip in stock price. Apples EU revenue is larger than China and Japan combined. Voluntarily forfeiting that is like the modern management equivalent of ritual self-sacrifice: afterwards, they would have so much unsold stock on their hands that it’s going to tank prices worldwide.
Meta is dead in the water with most gamers after they locked people out of their purchases and tried to shove a mandatory facebook login down everyone's throat.
Pimax is famous for over-promising and under-delivering - if they deliver at all, that is. Reddit is full of people complaining that they still haven't received their purchases from the launch before the last new announcement.
Apple’s Vision Pro was the price of a used car and you need to voluntarily lock yourself into Apple's ecosystem and pay 30% of all revenue to Apple for the privilege, so VR's biggest market (which is gaming) just said FY Apple and that's why it's one of the products with the highest return rate ever.
So you have Spyware vs. Vaporware vs. Masochism as your options. What a surprise that sales are down !!!! (not really)
I am in the market for a high-res VR headset for gaming. But there isn't any product for me to buy. Quest and Apple don't play well with PC. And for Pimax there's no shop actually selling them. (The article seems to agree, too. They have sales links to Meta and Apple, but no link for Pimax.)
They add to the robots.txt file:
"As a condition of accessing this website, you agree to abide by the following content signals ..."
which means robot operators are now better off if they never download the robots.txt file, because then you know for sure that you won't accidentally encounter these conditions.
This creates a legal risk if you try to adhere to the robots.txt, so it'll make future AI bots less controllable and less restricted.
That seems very reasonable. Otherwise, stealing money, lending it out for interest, and then returning the stolen money while keeping the interest would be a viable way to earn money.
"The cost of building a company on theft should be the company itself."
Also seems reasonable, because it is the same rule applied to AI companies.
I'd say leave 20% for the VCs and split the other 80% of Anthropic stock among the authors of affected books. That would surely discourage similar behavior in the future.
Thank you for sharing :)
"New online banking and new app
From 25 August 2025, you will benefit from the upgrade for online banking and Deutsche Bank app.
[..]
From 25 August, you will be able to simply reset your PIN yourself.
[..]
after logging in, you can also see accounts for which you are an authorised signatory."
But out of fairness, let me just mention that Chase behaves the same way. I think all of them just don't really care about small- and medium-sized businesses.
- from a number with no results on Kagi search
- claiming to be the online banking support of my bank
- asking me to read them a code sent to me via SMS
and when I refused to do that, they blocked my login credentials for online banking and sent me a sternly worded (paper) letter that my account could not be upgraded automatically for their software system migration because I had refused to engage with their support agent.
I then had to create a new login in their app, call the phone number on their letter and read that guy the SMS code and, to my surprise, that was the only !!! authentication needed to activate the new login credentials that I had just created.
(BTW, this was one of the top 100 largest banks worldwide)
It's almost like some companies are training you to fall for scams.
EDIT: This specific instance was Deutsche, but Chase has the exact same horrible habit of calling and then asking for an OTP code.
"Korean workers are being treated like criminals for building factories that Washington itself lobbied for. If this continues, investment in the US could be reconsidered."
"could be reconsidered"
In other words: The speculative decoding causes "holes" in your beam search data. You can fill them by sampling more, increasing hosting costs. Or you fill them with approximations, but that'll skew the results to be more "safe" => more generic, less reasoning.
"let's compare oranges to oranges"
That's impossible because Apple has bought up most of TSMC's 3nm production capacity. You could try to approximate by comparing Apple M4 Max against NVIDIA B300 but that'll be a very one-sided win for NVIDIA.
As someone who has implemented this myself, I know that it’s pretty easy to make innocent mistakes there. And the only visible result is a tiny distortion of the output distribution which only really becomes visible after analysing thousands of tokens. And I would assume that all providers are using speculative decoding by now because it’s the only way to have good inference speed at scale.
As a quick recap, you train a small model to quickly predict the easy tokens, like filler words, so that you can jump over them in the recurrent decoding loop. That way, a serial model can predict multiple tokens per invocation, thereby easily doubling throughput.
And the fact that they need lots of user tokens to verify that it works correctly would nicely explain why it took them a while to find and fix the issue.
And I believe that's also the real reason why Nintendo would block them: They make otherwise fluent games stutter, which will reflect badly on Nintendo and the game developers. Because people in general won't know that their Nreal Air is stealing some of the Switch's GPU performance and that they are creating performance issues themselves.
That means the monitor tries to change its own role from USB host (which it defaulted to because it's the power source) to USB gadget. And the Switch rejects the request because it sees itself as the USB gadget (for example to exchange data when you connect it to a PC).
The article itself also says "Our monitoring of the interactions with the USB-C monitor shows that the negotiation does not even get to the point of the Vendor Defined Messages(VDM)" so I see no hint of foul play by Nintendo. (And why would they? They are not in the business of selling USB monitors.)
My guess would be that the connection failure is due to a firmware bug in the monitor. But I am also heavily biased here because my last Asus screen had A LOT of firmware issues.
I feel like if this was heat related, the overall CPU temperature should still somewhat slowly creep up, thereby giving everything enough time for thermal throttling. But their discoloration sure looks like a thermal issue, so I wonder why the safety features of the CPU didn't catch this...