9 Comments
User's avatar
Neural Foundry's avatar

Fantastic retrospective on the predictions here. The adversarial perturbations trajectory esp caught my eye becuse it mirrors what I've seen in enterprise security audits this year. The licensing wave (Anthropic's $1.5B settlement) feels like a major inflection point, but dunno if AP adoption will scale fast enough before model providers patch the most egregious vulnerabilities. Working with a few infra teams recently, latent space reasoning seems to unlock way more efficiency gains than just throwing more compute at pre-training, especially for specialized domainstasks.

Devansh's avatar

Thank you

ToxSec's avatar

Would love to see an update on this for 2026!

Devansh's avatar

you’ll get it soon

Tiffany's avatar

It must be said that artificial intelligence has indeed brought many benefits to our lives!

Rainbow Roxy's avatar

Couldn't agree more. It's so insightful to see how your predictions playd out, a realy sharp analysis! Nexxa AI's work sounds fascinating, especially with the complex multi-agent computer use agents for industrial environments. Could you elaborate a bit more on the specific challenges these agents tackle in practice?

Devansh's avatar

A lot of stuff around dealing with legacy systems that don’t have data as readily available (hence computer use being a big thing).

Strongly recommend reaching out to Nexxa

Paul Gibbons's avatar

Such a lot of terrific work making some tricky content understandable!