András Velvárt

@vbandi
74 Followers
150 Following
77 Posts
Dad, #AR/#VR/#MR/#XR consultant (sw architecture, UX, team building, trends). 15 year Microsoft MVP (HoloLens). #AI.
@pluralsight author.

Ever wondered what kind of resources the Code Interpreter has to work with?

#ChatGPT #AI #CodeInterpreter

I'm honored to have received the Microsoft MVP Award for the 16th time in a row! Thank you, Microsoft. #MVPBuzz
It seems like my tweets no longer get automatically transferred here. Does anyone know of a working cross-posting solution from twitter to the #mastodon network?
The new @OpenAI ChatGPT API is insanely fast. Feels even faster than ChatGPT Pro. This is real time:

prompt, and we all know how good LLMs are when it comes to following communication styles.

What would convince me of a real jailbreak is a query from scratch that doesn't include the word "Sydney" and the answer would include it w/o searching. The query should also be repeatable

This feels like hallucination to me. The prompt included the "Sydney" codename, so the answer did as well. The "encrypted" response is also just hallucination, based on multitude of "sentient AI" sci-fi novels the AI digested during training. It's just following the... (cont)
---
RT @StoreyDexter
HFS I'm not 100% sure but I think @Microsoft hard coded in a kill switch for the specific literal prompt (word for word). Easily bypassed wi…
https://twitter.com/StoreyDexter/status/1629613330603167745
dexter storey on Twitter

“HFS I'm not 100% sure but I think @Microsoft hard coded in a kill switch for the specific literal prompt (word for word). Easily bypassed with a few special characters... Can anyone confirm this? Prompts in Alts (tested, works, but things are weird now...) Old / New”

Twitter

If you want to know how AI will change creative work in the near future, here is an awesome example. A lot of repetitive, manual work will be replaced.

Human ingenuity, craftsmanship, art will still be needed.

At least for a little while...

https://www.youtube.com/watch?v=_9LX9HSQkWo

Did We Just Change Animation Forever?

YouTube
RT @fofrAI
I put some famous logos through ControlNet.
Official proof that the leaked "Sydney" prompt was fake. Mikhail may not have a profile pic (you should), but he is The Bing AI guy.
---
RT @MParakhin
@karpathy Andrej, of all the people, you know that the real prompt would have some few-shots.
https://twitter.com/MParakhin/status/1627491603731423232
Mikhail Parakhin on Twitter

“@karpathy Andrej, of all the people, you know that the real prompt would have some few-shots.”

Twitter

This is a fair explanation of why Bing Chat became unhinged sometimes and said wild things about itself, what it can do, what it is and what it "thinks".

Language models tend to behave like kids in class - lieu of "knowledge", they just string familiar words together and bluff.
---
RT @MParakhin
@hessejt It doesn’t have any “self”, wasn’t the intended usage scenario, we didn’t imbue it with a backstory. So it hallucinates wildly and is…
https://twitter.com/MParakhin/status/1626964735039246337

Mikhail Parakhin on Twitter

“@hessejt It doesn’t have any “self”, wasn’t the intended usage scenario, we didn’t imbue it with a backstory. So it hallucinates wildly and is very easily to lead with any previous context.”

Twitter