both Google and OpenAI are shipping furiously

We should continue to take with a large shaker of salt any optimistic studies from the big players. They have a vested interest in keeping the hype train going, and will probably try to leverage AI-centric YouTubers to be a part of that hype train.

People are still mostly in denial… AI tooling will get way better and much more integrated into most systems and existing tools during the next 5 years. Completely different world.

11:35 The dirty secret here is that AI will *not* have to achieve full parity with humans in tasks to be adopted to those tasks and have a huge economic impact. Getting to just 80% parity of AI agents/robots to human work plus intrinsic positives (cost, speed, rapid error recovery, can work around the clock, no human attitudes or emotional failings) will have a huge value proposition and fuel rapid adoption of those capabilities. We're seeing it already; Opus is killing it in this range, and GPT-5 is close.

randomly encountering people that are just talking to chat gpt outside like they would be talking to a person is wild

hmm, these tests seem to me to be like an advanced version of the imitation game from the Turing test. Cool to see that the idea still stands.

try out Emergent here: https://app.emergent.sh/landing/?via=wesroth Their mission is simple: help dreamers become builders. If you have that viral app idea, go check out Emergent.

20:30 the AI terms in this chain of thought here are very open to interpretation, so your reading of the chain of thought (or cot) may be that it wants to deceive the "watchers", but another reading may be that it is reminding itself to avoid deceiving the watchers, as it has been instructed to. I think that ambiguity in cot interpretation is what the paper was trying to address, not the ability of the model to engage in scheming as supposedly seen in the cot

"Stay frosty out there" 😁indeed

Love the Tentacle!

Great video Wes, thank you!