Feels like 5.5 turned Opus into an insecure bureaucrat. Attempts to achieve quality with endless reporting, rules, and checklists, but ends up killing momentum and innovation.
Meanwhile, Fable goes "gotcha boss", makes the necessary calls, and gets it 95% there in one go.
elum retweeted
Every religion in history has promised miracles and warned of the apocalypse. Anthropic is the first to file both with the SEC.
elum retweeted
The funny thing is everyone was saying AI slop code will cause a cyber crisis but it’s actually all the old human slop causing it and the AI is just cleaning it up
elum retweeted
I think AI risk advocates - especially those with vested interests - would have a lot more credibility if they presented their plans for mitigating existential risk in conjunction with plans for mitigating regulatory capture and wealth/power centralization risk.
eg if they sad, “here’s our plan for:
a) slowing down AI development AND
b) simultaneously we’re going to donate 49% of our (non-voting) equity to a public fund AND
c) here’s our plan for ensuring regulation doesn’t block healthy competition or unfairly advantage us commercially
When a) is presented without reference to b) and c) it becomes very hard to believe that the true motivation is an altruistic concern for public welfare.
elum retweeted
Great term.
Sending slop grenades at work is one of the fastest ways to my shit list.
Lazy work used to mean too little output.
Now, with AI, it often means too much and more work for everyone else.
@tobi call it "slop grenades."
A "Slop Grenade" is when you let AI produce the work and pass it on without adding any value (including checking it).
Someone else has to wade through it, catch the mistakes, and clean up the mess.
You save time and look productive but someone else pays for it.
I suspect that the probability-based foundation of LLMs pull them towards some statistical mean which is what we perceive as bland or cliche. Slop always existed, but LLMs seem compelled to promote it by design.
AI is extremely good at taking an existing song and producing it, but it cannot yet create a good song from a single prompt. It is very poor at writing non-cliche lyrics that accompany a melody - I think this is an edge case in AI that is especially hard to crack because although the individual layers of a song can be somewhat formulaic, the totality of a "good" song is rarely predictable.
elum retweeted
The tech companies built the internet, which took the advertising business away from newspapers, and then most of the news business, and the gatekeeping on the marketplace of public opinions. The papers see tech as the natural enemy
elum retweeted
The social media crackdown happens in the context of a regime that loses public support and is faced with subversion. Ending privacy and controlling opinions is the whole point.
elum retweeted
Before LLMs I believed the analogy—I think it was Yud’s—that making AI via chatbots was like making real flowers by getting really good at sculpting wax. For most of the past 25 years, I thought we’d hit RSI via RL before anything learned English.
In that regard, I see us as lucky. AGI will come from “thoughts” we can read, literally. As important, LLMs are an expensive industrial process, not conventional PC software. AI cannot trivially self-improve as it could were it mostly code—as many assumed it would be.
None of this was guaranteed. AGI was feared to happen in a basement. It was supposed to explode.
Because it doesn’t, we can let capabilities out one by one. We can see what genuinely sucks about AI. We can integrate, adjust, and live in the Kurzweilian line-fitting world.
Even if it’s been unclear at times how long our stay is, I’m grateful we’re here.
elum retweeted
Stop using the industry PR term “age verification” and start calling it what it is: ID checks for the internet
elum retweeted
It's disheartening to see so many root against Zcash.
We owe so much to the courage of the Zcash team: ten years ago, Zcash was a moonshot.
They were the 1st to deploy a *fully* anonymous payment system and the 1st to deploy a zkSNARK-based cryptosystem.
Rooting against the team being able to recover from this bug is rooting against all other serious privacy efforts for cryptocurrencies: they are all based on the same nullifier techniques via circuit-based zkSNARKs.
This is not a zero-sum game.
Zcash has been showing us the way for over a decade and will continue to do so.
e.g., Zcash will soon be the 1st team to deploy a *formally-verified* zkSNARK circuit implementation for anonymous payments.