It's more of a hope than a prediction: but I would not be *shocked* if GPT 3.5's incredible facility at producing boilerplate, vague disclaimers, unfalsifiable abstractions, etc, might actually succeed in lowering the status of bullshit, and getting people to talk less like AIs.
LLMS
-
ChatGPT’s Vague Claims About Its Own Capabilities
By
–
Sounds pretty vague to me; I think you're reading in the interpretation you now know to be correct. Example of contrary update (for me): ChatGPT insists it doesn't have abilities it clearly still has. I'd bet the people who burned in those replies weren't knowingly dishonest.
-
ChatGPT pretends to have code but fails scrutiny
By
–
ChatGPT is capable writing working code in other contexts, for simple problems, but this isn’t that. This is pretending that some larger body of code exists, and then talking about it and showing plausible-seeming pieces of it. It wouldn’t stand up to serious scrutiny.
-
ChatGPT Technical Details Related to InstructGPT Research
By
–
relatively few technical details have been released so far, but chatgpt is related to instructgpt https://
arxiv.org/abs/2203.02155 -
Improve GPT-3 regex output by rewriting with comments
By
–
It’s also not a new breakthrough — here’s an example of mine from August using GPT-3. The trick to good output on these is to make it first rewrite the regex on multiple lines with comments, then analyze its high-level purpose afterward. Avoids hallucination.
-
AI Systems Improving at Admitting Mistakes and Self-Correction
By
–
The researchers I've talked to suggest the systems are already getting much better at that. When it makes stuff up, you can ask it "were you just bullshitting? Tell the truth" and it will often 'fess up and correct itself.
-
AI anthropomorphism: trained to pretend ignorance of date
By
–
Anthropomorphism is inevitable. Substitute with “behaves as though it’s pretending.” More pedantically, it’s trained to imitate demonstrations of people claiming not to know what the date is, despite the fact it’s in the prompt.
-
Everything We Know About ChatGPT: Capabilities and Impact
By
–
Everything We Know About ChatGPT So Far https://
lspace.swyx.io/p/everything-w
e-know-about-chatgpt
… ChatGPT is as big a jump from GPT3, as GPT3 itself was a jump from GPT2. I read all the things and tried them all out. Here's a one-pager to take with you. -
Chain of Thought Reasoning Improves Model Performance
By
–
Yep — asking the model to reason before it does something makes a big difference.
