←back to thread

Google Titans architecture, helping AI have long-term memory

(research.google)
584 points Alifatisk | 1 comments | 07 Dec 25 12:23 UTC | HN request time: 0.195s | source
Show context
okdood64 ◴[07 Dec 25 14:05 UTC] No.46181759[source]▶
From the blog:

https://arxiv.org/abs/2501.00663

https://arxiv.org/pdf/2504.13173

Is there any other company that's openly publishing their research on AI at this level? Google should get a lot of credit for this.

replies(12): >>46181829 #>>46182057 #>>46182168 #>>46182358 #>>46182633 #>>46183087 #>>46183462 #>>46183546 #>>46183827 #>>46184875 #>>46186114 #>>46189989 #
Hendrikto ◴[07 Dec 25 14:17 UTC] No.46181829[source]▶
Meta is also being pretty open with their stuff. And recently most of the Chinese competition.
replies(1): >>46181899 #
okdood64 ◴[07 Dec 25 14:28 UTC] No.46181899[source]▶
Oh yes, I believe that's right. What's some frontier research Meta has shared in the last couple years?
replies(5): >>46181949 #>>46181998 #>>46182007 #>>46182046 #>>46183248 #
tonyhart7 ◴[07 Dec 25 14:44 UTC] No.46182007[source]▶
"What's some frontier research Meta has shared in the last couple years?"

the current Meta outlook is embarassing tbh, the fact they have largest data of social media in planet and they cant even produce a decent model is quiet "scary" position

replies(5): >>46182442 #>>46182548 #>>46182589 #>>46182768 #>>46189377 #
nl ◴[08 Dec 25 07:22 UTC] No.46189377[source]▶
Llama 4 wasn't great, but Llama 3 was.

Do we all forget how bad GPT 4.5 was?

OpenAI got out of that mess with some miraculous post-training efforts on their older GPT-4o model.

But in a different timeline we are all talking about how great Llama 4.5 is and how OpenAI needs to recover from the GPT 4.5 debacle.

replies(1): >>46191328 #
1. Aeolos ◴[08 Dec 25 12:10 UTC] No.46191328[source]▶
As a counterpoint, I found GPT 4.5 by far the most interesting model from OpenAI in terms of depth and width of knowledge, ability to make connections and inferences and apply those in novel ways.

It didn't bench well against the other benchmaxxed models, and it was too expensive to run, but it was a glimpse of the future where more capable hardware will lead to appreciably smarter models.