Was running benchmarks on Minimax H3 locally and one of the random clips it spit out clearly exposes the company for having trained their model on the film POOTIE TANG. Amirite?
Good advice.
Like you're 5: Lots of new shops only rent fancy AI computers. Banks stopped lending them money unless they already earn from other stuff. Strong ones like SpaceX (rockets + Starlink) still get loans. Most of the 200+ little shops will close or merge. Few survive.
Prepare: Lean into your local DGX. Master efficient home models. Avoid relying on any one cloud.
Income: Sell private local video gen/automations. Help others set up home AI. Build offline tools as clouds consolidate.
The @ArtificialAnlys rankings of the top models. If accurate, this new addition is significant and indicative of a growing potential business concern / trend for certain frontier labs. Yes?
Jace Hall retweeted
I grew up on arcade games. This weekend, I built a haunted one. 🎃🍭👻
Trick or Treat Eat Eat: grab treats, gulp ghosts, chase an S rank.
Built with Rive CLI + GPU Canvas.
Play: trick-or-treat-eat-eat.verce…
@rive_app #rivehalloweenchallenge
Jace Hall retweeted
Gods Don't Give Gifts
Coming to theaters December 4th.
On Niiro Cradle, where bodies are disposable and consciousness has been commoditized, four interconnected stories collide with the horrifying consequences of humanity’s attempt to transcend the flesh.
Jace Hall retweeted
Poker tournaments are intimidating, especially if you’ve never played one. I wanted to make a game that breaks down those barriers and makes poker more approachable and fun. If you’ve ever been afraid to try poker, this game is for you. Currently in Beta.
Aceseveryhand.com
When Doc Brown screamed "1.21 gigawatts!" in 1985, it was meant to sound like an impossibly sci-fi amount of energy. Today, 1.21 gigawatts is the exact scale of a single ultra-massive AI datacenter campus.
The primary difference is that the DeLorean needed 1.21 gigawatts in a single instantaneous burst, whereas modern AI facilities consume this jaw-dropping amount of electricity continuously, 24/7.
There used to be an assumption built into a long message from someone you respected.
If an expert sent you 800 words, you generally assumed they had already thrown away the other 2,000. The length was the result of judgment.
AI has quietly broken that assumption.
Now, if a message feels AI-assisted, people tend to assume the opposite. They assume they are seeing whatever the machine produced, more or less unfiltered, unless there is some obvious sign that a human actually curated it.
And the main sign we seem to accept as proof of curation is that it got shorter.
That is a strange change.
Length used to imply effort. Now length can imply the absence of effort.
Which means brevity is starting to function as evidence that a human was involved.
The part that bothers me is what this reinforces.
We already live in a culture that has become far too comfortable treating simple, short, confident statements as more credible than complicated ones. You can see it everywhere in media, politics, social media, basically any place where ideas have to compete for attention.
But reality is not obligated to be simple.
Sometimes the accurate answer is long because the thing itself is complicated.
So if our suspicion of AI causes us to increasingly equate compression with human judgment, we may accidentally strengthen an already terrible heuristic:
Shorter means more considered.
Simpler means more credible.
Therefore simpler is probably more true.
Which is obviously nonsense.
Anyway, you probably stopped reading three paragraphs ago.
This was way too long.
So I’m probably wrong. 👀
I work so near-continuously with all the frontier model simultaneously that any subtle change that takes place with them operationally (guardrails, filters, etc.) I almost immediately feel the effects of.
Gemini just went through one. Let's see how long it takes people to notice or if google even mentions it.
