It *could* technically be unrecognisable blend of faces with infinitesimally small contributions from the input data. But it would look average and suspicious…
SAFETY
-
Why Fictional AI Differs Radically From Real AI Systems
By
–
The fact that our current form of AI is so different from the AIs in science fiction blinds our imagination. In fiction, AIs are cold calculating machines driven by logic (unless you make them go insane by giving them a paradox). We have warm, weird, fallible, gullible systems.
-
Training AI to Forget and Forgive Optimally
By
–
It should be easy enough to train an AI to also forget and forgive. Maybe there's even an "optimal" amount of forgetting that we should all aim for, perhaps with help.
-
AI Likeness Theft: Why Real Data Beats Pure Generation
By
–
To replies asking: "Why did they steal her likeness rather than generating someone from scratch?"
— Alex J. Champandard 🌱 (@alexjc) 9 mars 2024
Because AI alone does not breach the uncanny valley. It has to be example-based to be anywhere close to trustworthy. (Ab)using real people and real content will remain necessary. https://t.co/LnUutk8n2yTo replies asking: "Why did they steal her likeness rather than generating someone from scratch?" Because AI alone does not breach the uncanny valley. It has to be example-based to be anywhere close to trustworthy. (Ab)using real people and real content will remain necessary.
-
Gemini Privacy Warning Regarding Human Review of Conversations
By
–
Jokes aside, now Gemini warns you not to write anything sensitive or private because your conversations may be reviewed by real people.
-
Claude Identifies Georgetown House: Real Understanding or Hallucination?
By
–
Example of why working with AI both so impressive and so challenging: I show Claude a picture of a house in Georgetown that shouldn't be in its training set. It nails it (GPT-4 does too) I ask it why Georgetown? Its answers seem great, but could all be hallucinated justification
-
API Base Models Fine-tuning and Guardrail Instructions
By
–
As far as I know, you are not getting the base model through APIs, there is definitely fine-tuning and many guardrail instructions that are likely the result of prompting. Refusals to mess with copyright work, for example.
-
GPT-4 Benchmark Bias: Circular Evaluation Problem
By
–
I could not agree more, to the point that I have lost trust in most of these benchmarks. Plus, if GPT-4 is used to run the evals, isn't that an inherent bias of the system?
-
Google AI Secrets Stolen: Top AI Stories Today
By
–
Top stories in AI today: -Google engineer steals AI secrets
-Inflection upgrade nears GPT-4
-Generate an AI song with just a prompt
-Researchers create self-spreading AI malware
-6 new AI tools & 4 new AI jobs Read more: http://
therundown.ai/p/googles-ai-s
ecrets-stolen
… -
Neural Networks Struggle with First-Order Logic Learning
By
–
As pointed out by @sirbayes
, this paper https://
starai.cs.ucla.edu/papers/ZhangIJ
CAI23.pdf
… has a formal investigation into the observation from the first tweet — that "my models were completely unable to learn to perform actual first-order logic — despite the fact that this ability was definitely part of