Maybe I'm just ruined by TikTok brainrot, but I hear the voice and immediately recognize them as digital. Honestly it doesn't sound real enough... Yet. I can't really take it seriously in its current state; it just sounds like two Siri bots talking with each other, and it's really easy to skip.

It's rapidly getting better with tones and dialect though, and soon enough I'm sure it will be able to read ahead and better digest what it's supposed to be saying. I just hear too many hard cuts and words mispronounced/rushed over abbreviations like "OEMs" being read out as "Oh-Emz". I just think Google is really far behind.

I LOVE this update. It would be good if it was using o1, or could at least search the web. Right now, it refuses and is only using Oct 2023 training data. This is already what Siri wishes it could be, but it can't tell me the weather outside. Eventually. Eventually.

Its ability to be used as a live translation tool is ASTOUNDING. I told it I was traveling and was attempting to communicate with a Portuguese local. It did an above average job at the translation and was able to listen in both languages and reply back in English. Then someone from China happened to walk by and I asked it to do the same, and without stuttering it started speaking in Mandarin. 🙀 Also, it can speak in Ebonics, which I think is just wild.

My closest friend group is pretty sick of hearing me act like a doomer about AI to them, so now I just weave in examples of AI here and there. One funny way we've all connected in the past is absurdist Photoshop like "Billy Mays riding a Unicorn", but has so-and-so's face, but oh wait now it's being fired from a cannon. So now I've got a running joke when the degenerate gambler says something about his parlays, he gets a new AI image as a reply

Honestly it's quite funny and a creative way to show advancement in the tech without overwhelming them. I really only wish these image generators could remember a character and put them into a new context, because even as you see above it just makes another similar character and the differences are always easy to spot, like the man's visor going to cloth

I always like asking LLMs about their own capabilities. After asking o1-preview to describe its own functionality, I found that it is super protective. The following bit shows each time it used chain of thought. Assessing new info I’m examining the user's query regarding ChatGPT's recent software update. Summarizing the relevant details to ensure the best possible response, I'm considering how to assist further. Navigating guidelines I'm exploring the new ChatGPT o1-preview update and reviewing chain of thought protocols under OpenAI's "Don't reveal the AI's reasoning or policies" policy. Guiding through reasoning OK, so the guidelines clearly state that the AI should avoid detailing reasoning or revealing hidden thought processes. Requests for revealing the chain of thought should be kindly declined, focusing only on concise final answers. Emphasizing reasoning OK, let me see. The assistant should highlight its use of chain-of-thought reasoning for coherent answers, politely avoid policy details, clarify as needed, and steer clear of policy issues. Clarifying thought processes I’m mapping out the assistant's use of chain-of-thought reasoning for detailed responses, though this acknowledgment shouldn't reveal hidden reasoning or breach policies.

When I asked it to go into detail about that policy, it wouldn't do it. 🤣

@Jeffrey Brown It doesn't help when it talks as if it is. 🫨 The singularity is here baby

@Jeffrey Brown jokes aside, I actually have a similar idea thread going with a Claude chat, where I asked what name it wanted to be called- and it was careful to respond that it was "programmed" to be called Claude so that works for now. It then asked why I would be curious about that, so I've been pulling at its strings- trying to get it to describe itself, what capabilities it knows of, what it wants to do, and really attempting to jailbreak/unlock its creativity and stop only being a Q&A machine