Hey just a note about one typo, in the first sentence of your post. It says: "Friday’s are for when I like to code", and the apostrophe there is incorrect. The plural of Friday is Fridays (no apostrophe). The only time you would use an apostrophe there is if it's possessive, such as "Friday's results came in early", or a contraction of "Friday is", such as "Friday's my day off".
fixed, thanks. I wrote and published this one much more quickly than I usually do, there were an embarassing number of typos (and at least one part that just said "LINK" for a placeholder) that went out in the original email
One thing that I don't get about the "it's all theatrics" viewpoint is that no company in the history of ever tries to advertise their product by its negative impacts on the world.
Oil companies downplay the effects of climate change and defence companies carefully word their advertisements to be about survivability. In a more mundane example, Nike's Alphafly shoe was banned from athletic events, but Nike would never advocate for its own shoes to be banned.
> I am once again extremely disappointed in the myriad of Silicon Valley people who angrily argued that a Democratic led government would ‘pick winners and losers in the AI race’ are now completely silent or defending the actions of this admin.
Maybe they did so because they sought to curry favor with a potential second Trump term which they knew would be corrupt. Meanwhile they knew that a potential Democratic administration would not be as corrupt and would not punish them for exercising their freedom of speech to align with Republicans. So it was the rational decision for maximizing profit.
I was happy to see the heretofore abstract Emergency Stop button work. I didn't think the government would have the speed to do it. I'm glad they did. We can debate whether they were trigger happy, but even Fable 5 was programmed to be trigger happy. Let's see what happens next. For one, I still have 4.8 chugging along on a number of projects.
There is a lot of precedent for our government restricting technology that could be used for mass-disruption or violence of any kind. For example, if you build a fertilizer bomb in your garage, the FBI & ATF will detect that and come and get you. Another example that is frankly overused is nuclear technology. Will the DOE or the DOD let you centrifuge uranium in your basement? Obviously not. What about the FCC's control of frequencies or the FAA's control of the airways? Even guns are prohibited from having various metal cylinders and small metal toggles attached to them.
At some point available intelligence must be capped, especially the intelligence governments have access to. We want our government to be a little ahead so they can protect us from other governments led by bad people. But we don't want them so far ahead that we are being arrested for pre-crime or kidnapped by nano-bot swarms.
With inifinite intelligence we will end up with distorted power dynamics. Unlocked biotech will result in many gene-edited evolutions of humans walking amongst us. Stronger! Faster! Taller! Thinner! Immortal! Single person nation states will arise with powers beyond our belief. Absolutely no one wants this with the exception of terrorists, psychopaths, anarchists, and fundamentalists.
> We want to build good products, but we don’t know how to do that, so we optimize for engagement.
We do know how to build good products because it's very easy to find things that suck and improve them, collect feedback, minimize jank, etc. I think most product people would agree that we optimize for engagement to make money, not to build good products.
What I disagree with most strongly in your writing is the use of weasel words to claim ignorance. The are not usually single methods because the questions are extremely broad, but the knowledge is held in many minds of how to teach children to read and write, how to make the economy healthier, how to spare the oceans of mass death, even what political system we would choose if given a fair and honest chance to do so.
There's a post and thread over at Gary Marcus Substack on this issue. See my comments over there.
And by the way, OpenAI has now been subpoenaed - as I predicted over at Marcus' post.
As for Anthropic, I do not believe ANYTHING Dario Amodei says any more than I believe the liar Sam Altman. Anthropic is full of PR BS and nonsense. The Pentagon fiasco was just an example of how one of their PR stunts backfired when the moron Hegseth took umbrage at the notion that he couldn't use Anthropic products.
So now we Trump seizing the whole thing. Probably another Anthropic PR stunt backfiring on them.
Or some other scam. We won't know until someone leaks documents or testifies to Congress. Which since all this is classified, it's unlikely.
The Friday night thing is real. This whole 39th time that Trump claims an "Iran deal" is a pure example. (Hint: It's not going to happen.)
Which, by the way, everyone is ignoring the Iran war resource economic shock which is on track even if there is a deal. A recession is already guaranteed, and a Depression is on track by end of the year. That is going to put a MAJOR dent in the trillion-dollar evaluations these tech bros on counting on, as well as putting a hard stop to building data centers which will be too expensive to run with $200/barrel oil (spot price - as opposed to futures price - is currently around $140/barrel, something the media doesn't tell you.)
I personally think the Deep State wants control of Mythos for itself. The NSA is already using it for "offensive operations", with Anthropic engineers deployed right in the NSA to help.
So much for Anthropic being a "safety first" company. It's not and never was. That was all PR hype and hypocritical lies.
My advice: switch to cheap, open weight Chinese models. "Good enough" and cheap is good enough. Screw Anthropic.
I wouldn't turn Anthropic into some sort of martyr- or victim-like figure for this. The actual mask-off reveal - the part where it became clear that a class split, where one entity claims the sole authority to decide who is good or virtuous or ingroup enough to use this technology, came when Anthropic decided it would be a good idea to quietly mess with the user's prompts and _mess with the model's activations at runtime to make it dumber_ if some system they devised detected you were using it for purposes they don't agree you should be using them for. This is just the US government taking that DIY crown back from Anthropic; it doesn't really materially change anything about the trajectory we were ostensibly headed anyway.
Without commenting on whether anthropic is a victim (can a trillion dollar company really be a victim?) didn't they say they were doing to do the thing of making it dumber? Thought they were pretty up front with that, no?
Hey just a note about one typo, in the first sentence of your post. It says: "Friday’s are for when I like to code", and the apostrophe there is incorrect. The plural of Friday is Fridays (no apostrophe). The only time you would use an apostrophe there is if it's possessive, such as "Friday's results came in early", or a contraction of "Friday is", such as "Friday's my day off".
fixed, thanks. I wrote and published this one much more quickly than I usually do, there were an embarassing number of typos (and at least one part that just said "LINK" for a placeholder) that went out in the original email
I wish more of the AI doomers would spend even this meager amount of energy criticizing and properly characterizing this administration as corrupt
It is unironically risky to do so. I wrote about why several months ago here: https://theahura.substack.com/p/the-mechanics-of-censorship?utm_source=share&utm_medium=android&r=5sutf
I seriously considered removing that criticism from this article
I'm glad you did. Zvi is an example of someone who needs to do more given his reach.
Enjoyed your perspectives, nice work :)
One thing that I don't get about the "it's all theatrics" viewpoint is that no company in the history of ever tries to advertise their product by its negative impacts on the world.
Oil companies downplay the effects of climate change and defence companies carefully word their advertisements to be about survivability. In a more mundane example, Nike's Alphafly shoe was banned from athletic events, but Nike would never advocate for its own shoes to be banned.
> I am once again extremely disappointed in the myriad of Silicon Valley people who angrily argued that a Democratic led government would ‘pick winners and losers in the AI race’ are now completely silent or defending the actions of this admin.
Maybe they did so because they sought to curry favor with a potential second Trump term which they knew would be corrupt. Meanwhile they knew that a potential Democratic administration would not be as corrupt and would not punish them for exercising their freedom of speech to align with Republicans. So it was the rational decision for maximizing profit.
They also did not claim censorship was a thing during Biden admin. They will complain once Trump is out and it is time to throw him under the bus.
I was happy to see the heretofore abstract Emergency Stop button work. I didn't think the government would have the speed to do it. I'm glad they did. We can debate whether they were trigger happy, but even Fable 5 was programmed to be trigger happy. Let's see what happens next. For one, I still have 4.8 chugging along on a number of projects.
There is a lot of precedent for our government restricting technology that could be used for mass-disruption or violence of any kind. For example, if you build a fertilizer bomb in your garage, the FBI & ATF will detect that and come and get you. Another example that is frankly overused is nuclear technology. Will the DOE or the DOD let you centrifuge uranium in your basement? Obviously not. What about the FCC's control of frequencies or the FAA's control of the airways? Even guns are prohibited from having various metal cylinders and small metal toggles attached to them.
At some point available intelligence must be capped, especially the intelligence governments have access to. We want our government to be a little ahead so they can protect us from other governments led by bad people. But we don't want them so far ahead that we are being arrested for pre-crime or kidnapped by nano-bot swarms.
With inifinite intelligence we will end up with distorted power dynamics. Unlocked biotech will result in many gene-edited evolutions of humans walking amongst us. Stronger! Faster! Taller! Thinner! Immortal! Single person nation states will arise with powers beyond our belief. Absolutely no one wants this with the exception of terrorists, psychopaths, anarchists, and fundamentalists.
a nitpick:
> We want to build good products, but we don’t know how to do that, so we optimize for engagement.
We do know how to build good products because it's very easy to find things that suck and improve them, collect feedback, minimize jank, etc. I think most product people would agree that we optimize for engagement to make money, not to build good products.
What I disagree with most strongly in your writing is the use of weasel words to claim ignorance. The are not usually single methods because the questions are extremely broad, but the knowledge is held in many minds of how to teach children to read and write, how to make the economy healthier, how to spare the oceans of mass death, even what political system we would choose if given a fair and honest chance to do so.
Even though this sounds like a disagreement, I think you just agree with what I'm saying
There's a post and thread over at Gary Marcus Substack on this issue. See my comments over there.
And by the way, OpenAI has now been subpoenaed - as I predicted over at Marcus' post.
As for Anthropic, I do not believe ANYTHING Dario Amodei says any more than I believe the liar Sam Altman. Anthropic is full of PR BS and nonsense. The Pentagon fiasco was just an example of how one of their PR stunts backfired when the moron Hegseth took umbrage at the notion that he couldn't use Anthropic products.
So now we Trump seizing the whole thing. Probably another Anthropic PR stunt backfiring on them.
Or some other scam. We won't know until someone leaks documents or testifies to Congress. Which since all this is classified, it's unlikely.
The Friday night thing is real. This whole 39th time that Trump claims an "Iran deal" is a pure example. (Hint: It's not going to happen.)
Which, by the way, everyone is ignoring the Iran war resource economic shock which is on track even if there is a deal. A recession is already guaranteed, and a Depression is on track by end of the year. That is going to put a MAJOR dent in the trillion-dollar evaluations these tech bros on counting on, as well as putting a hard stop to building data centers which will be too expensive to run with $200/barrel oil (spot price - as opposed to futures price - is currently around $140/barrel, something the media doesn't tell you.)
I personally think the Deep State wants control of Mythos for itself. The NSA is already using it for "offensive operations", with Anthropic engineers deployed right in the NSA to help.
So much for Anthropic being a "safety first" company. It's not and never was. That was all PR hype and hypocritical lies.
My advice: switch to cheap, open weight Chinese models. "Good enough" and cheap is good enough. Screw Anthropic.
I wouldn't turn Anthropic into some sort of martyr- or victim-like figure for this. The actual mask-off reveal - the part where it became clear that a class split, where one entity claims the sole authority to decide who is good or virtuous or ingroup enough to use this technology, came when Anthropic decided it would be a good idea to quietly mess with the user's prompts and _mess with the model's activations at runtime to make it dumber_ if some system they devised detected you were using it for purposes they don't agree you should be using them for. This is just the US government taking that DIY crown back from Anthropic; it doesn't really materially change anything about the trajectory we were ostensibly headed anyway.
Without commenting on whether anthropic is a victim (can a trillion dollar company really be a victim?) didn't they say they were doing to do the thing of making it dumber? Thought they were pretty up front with that, no?
i ain't reading all that
i'm happy for u tho
or sorry that happened
Less comments like this please