Remarkably brief.

There's possibly the USVI tax thing as well but I don't know if he used that

One weird thing about social media, is that on xᵀwiᵀᵀer accounts still begin a dialogue with something that they could have easily asked an llm

Here let me Gemini that for you

Most interactions are mogged by a good use of an llm, fine, I'd rather have a higher bar for interactions

But he said "quantitative trading firm" and that has some degree of specificity. Though precision in language among other tech opinion havers is them saying "alpha" when they mean "edge"

Also no hate for this guy at all. But it's my opinion on an opinion

Unless he means fake returns which is probably one thing an AGI would do, or illegal returns which is a much more achievable as reasonable thing for AGI to do

Everything I think I know about a topic I think I know a little bit about says this is very wrong. And it is very weird to me considering who it is

Claude is this true? Think hard

Got that Opus 4.6 showing up but but it comes with a warning label so I'm afraid to use it

Is $X a historically good ticker?

McAfee was much more interesting than Saylor. Just the worst people around now

Deception

Violation

Treachery

Betrayal

Freaking Anthropic, do they know customer retention is a thing

People shouldn't gamble. But there are a lot of shouldn'ts and I'm not going to speculate on which are the most problematic

That mNAV metric is kinda dumb and annoying. Wish I had never read about it

Presumably rational people are using prediction markets for risk management so the losses are offset by gains elsewhere. Other people are just simple salt of the earth utility maximization enjoyers.

The argument, motivated by a jaundiced eye, is that some people are so powerful, so evil—
the puppeteers and masterminds —they puposely avoided being in the file. Absence is suspicious because, according to these nutters, the big evil team should all be friends. Typical conspiracy slop.

"What have you done with my monkeys‽"

I could communicate with inline stickers to ensure humanity, for example the message in the photo. It turns out an LLM can read and paraphrase this perfectly. I wonder how difficult to replicate.

I could imagine people trying to use neologisms or newly minted phrases, post knowledge cutoff, to signal humanity. Maybe then language evolves-—-well—changes, quickly and that could go either way but I'm not optimistic.

Style, is maybe more subjective. All caps—gross. Using '=/=' is also visually disturbing and a poor choice when '!=' or '≠' are available

A problem with anti AI sentiment and also a problem with AI model optimism, is human slop.

I feel like it is more frequent now in human writing.

For example, the phrase "many such cases" is, for me grating human slop. This is just one example there are…m--others.

Maybe some types of benchmaxxing are better than others, say code, because—possibly—good code taste and development are better aligned. Though coding primates on social media seem to be a heterogeneous bunch

The thing that could happen is that there aren't many labs, there are enough people with poor taste that the less fit models+ win—we get the VHS model. In my imagination the bad outcome happens because the people best suited to curate and provide feedback are not hired by the labs or given agency

One thing that I would like to believe—but don't know if true—is that good future AI models (and the benchmark is relevant) will benefit from human labor: the good taste of humans developing them.
I guess that could be in model development, scaffolding, evaluation (includes the benchmark).

Oh and another DeepSeek model in a couple of weeks. But nothing ever happens

I see we are doing $BTCQ again

I just saw a thread where everyone mostly agreed that not being in the files was even worse and totally suspicious—though somewhat contingent on who.