this post was submitted on 24 Mar 2026
166 points (97.2% liked)
Technology
83032 readers
4100 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
I think it was build on the original transformer architecture and as such took a shitload of compute and was slooow, so I guess they picked the wrong architecture and had to scrub the whole strategy. Huge loss.. That also point to US not having enough cheap compute (compute->combined mem and processing) - likely from missing electricity. Lovely.. Die Saltman, die, and Go go China !
Bops the tankie.
Like, I have a Chinese LLM loaded right this second and follow them closely, but holy moly. Curb your enthusiasm.
Anyway, OpenAI has plenty of compute to train a Sora 2 if they want, but apparently they don't. My guess is some combination of:
They couldn't figure out a more efficient architecture, like you speculated. I buy that. OpenAI's development is way more conservative than you'd think, and video generation is inherently intense, especially if Sora 1 is the baseline.
...Maybe they looked at metrics, saw Sora is mostly used for spam, scams, or worse, and pulled the plug for liability reasons?
They're focusing on short-term profitability, as other commenters mentioned.