Is it okay to be amazed?
Posts by Benjamin
I posted my Senate testimony about Section 230 and responses to follow-up Questions for the Record on SSRN, and wrote this blog post excerpting the main thrust of the testimony.
It describes the positive work Section 230 is doing today, and what we know about *actual outcomes* under alternate laws.
My point was more that you are playing his game. Arguing against him is like posting your wordle solution.
You may think of it as a win, but he only sees the traffic you drive to his site.
If you're having fun, great. If you want him to go away, don't give him clicks.
Ed Zitron is a "public relations specialist". He's published books on it. He has no technical or financial background.
He skillfully makes us talk about him and makes money with that.
Saying that he is wrong is like saying sci-fi science is wrong. That's just not the point.
Fair use isn't about using machines or having resources. It's not a loophole either. The legality of AI training follows straight from the copyright clause in the US Constitution. It's absolutely fundamental.
Europe sees this differently, but not in a way I'd call reasonable.
AI training is not the same as removing all IP laws.
It's imo scarier we're not finding giga-brained bugs yet, boundary pushing helps bound your "unknown unknowns", human-tier attackers, and weaker-LLM script kiddie attacks.
Still just finding "haha oopsies!" bugs means you're safe from neither the fruits of incompetence nor the beasts in the forest.
What about that the fact that these could have been found at any time in the past is encouraging?
How so?
According to the US Constitution, patents and copyrights are supposed to encourage engineering and science.
That's ultimately why AI training is Fair Use in the US.
Europe is taking a different approach. It's quite instructive.
Excited to launch the accompanying free RLHF Course for my book. To kick it off, I've released:
- Welcome video
- Lecture 1: Overview of RLHF & Post-training
- Lecture 2: IFT, Reward Models, Rejection Sampling
- Lecture 3: RL Math
- Lecture 4: RL Implementation
Landing page: rlhfbook.com/course
Worker-owned co-op? You mean like a tech start-up?
The big problem with such ideas is that fundamental economic tasks still need to be done.
You could look at attempts to lend money without charging interest. That was/is regarded as usury in some religions.
Yes, but they didn't ask about their own experience. Frequent AI users are more positive, but causality is unclear. It correlates with parental use, though ...
I clicked through to the report for the details.
Still not AGI. AGI would have roasted you for writing "much" instead of "many".
The ethnic breakdown is fascinating. White people are the most negative, which I interpret as status fears.
Black people are most hopeful about the impacts on learning. Maybe indicates lower quality of education offered to that demographic?
There is no real indication that views were changed by their own experience, rather than by what parents or media told them. Shame.
This data on US Gen Z (age 14-29) attitudes to GenAI is pretty interesting. A real increase in anger/anxiety in 2026 compared to 2025. And then a lot of smaller but consistent shifts to more negativity about how it actually helps them work or learn.
news.gallup.com/poll/708224/...
"It is not just drag shows that are placed at risk by these politicians. In its puritanical zeal to root out new forms of 'indecency,' the legislation even bans sports bras and other kinds of women’s clothing when worn in 'physical proximity' to those who are not members of the woman’s household."
An AI that can generate eg images can generate defamatory images. There is no engineering solution that isn't simply surveillance. Perhaps RFC 3514 needs updating.
The point was that a vehicle is heavy machinery. Its purpose gives it inherent destructive potential.
Engineers have come up with ways to make cars safer. But it is not possible to comply with a law demanding that cars should be absolutely safe.
And that's terrible news for comedians, because no new material is needed. Maybe comedy writers could retrain as smelters?
www.youtube.com/watch?v=nyu4...
of thing before publication is totalitarian surveillance. One must monitor all computers and even phones to catch it.
One doesn't have to invoke chilling effects on speech to reject such ideas.
I'm hard-pressed to come up with any desirable AI-regulation that would be prevented by 1A.
Demanding a specific ideological bent for models is not desirable.
An AI that can't make defamatory outputs (eg sexual deepfakes) is like a car that can't run over people. The only way to stop that sort...
That is socialism according to Lenin and Stalin.
"He who does not work, neither shall he eat." That's from the bible but also found in the 1936 Stalin constitution of the Soviet Union.
The UBI is one of the few things that are neither left nor right but liberal.
en.wikipedia.org/wiki/He_who_...
Huh? How does their versioning work?
Wouldn't the harness have to include a chemical lab for that?
The Tech Outrage Playbook wins again:
alecmuffett.com/article/139710
4 frame comic. first frame shows a swimmer with a one piece bathing suit. 2nd frame shows an enthusiastic kid with a shirt written waterpolo on it. 3rd shows a man with a shirt that reads "someone I love is trans". 4th frame shows a girl with a trans flag. text reads, across the frames : Being visible or outspoken as a trans person will make you a target of violence and hatred. but it will also show every other trans person that it's possible to exist in their space. visibility empowers and changes heart. I believe in a future in which trans people aren't forced to live in fear, shame and secrecy.
"I believe in a future in which trans people aren't forced to live in fear, shame and secrecy."
My strip for TDOV.
Be your own basilisk.