Copilot also has lots of grounding in the codebase through clever prompt engineering so that might be what you’re noticing.
@goodside
-
ChatGPT Hallucinations Reduced Compared to GPT-3
By
–
ChatGPT hallucinates a lot less. It makes stuff up in the details all the time, but normal GPT-3 will define nonsense words or imagine the text of fake URLs. See here for examples:
-
Perplexity.ai: grounded GPT-3.5 + Bing alternative to YouChat
By
–
Btw if you like the references-included style of YouChat above, also check out http://
Perplexity.ai (GPT-3.5 + Bing). No multi-turn chat, but the results are better grounded by the SERP and more useful overall. Great for summaries of recent controversies especially. -
YouChat combines ChatGPT and search, useful despite hallucinations
By
–
3. YouChat — http://
You.com ChatGPT + search Not super reliable — hallucinates often in spite of SERP grounding. But when it works, being able to ask conversational questions about recent, technical subjects is just incredible: -
Jasper Chat: Better prose than ChatGPT, but more hallucinations.
By
–
2. Jasper Chat — https://
jasper.ai/chat If you liked my posts on longer-form writing in ChatGPT using conversational feedback, this is what you want. Better prose than ChatGPT, and more imaginative. Fact-check hard, though — it hallucinates more too. -
Claim of ChatGPT political shift criticized for weak evidence
By
–
There's a popular thread claiming ChatGPT's political opinions shifted centrist after its Dec. 15 update. Evidence given for this is weak (4 online political tests given only once in each period), and the prompting method inappropriately forces in-test consistency of answers:
-
Galactica less popular, more hallucinatory than ChatGPT
By
–
Less, I think. Galactica didn't land with the consumer-facing splash that ChatGPT did. I only used it briefly, but it was also more hallucinatory, e.g. never refusing absurd requests. I think hallucination can be worse when infrequent, because we trust the model more.
-
Optimistic about LLMs but ChatGPT replies still suck
By
–
For those who don't know me — I'm very optimistic about LLMs in general. The visible problems of today's LLMs will be fixed, and long-form LLM writing is obviously useful to many people today. At the same time, I think StackOverflow and HN have a point. ChatGPT replies suck.
-
NYT think-pieces on LLM social faux pas expected by January
By
–
I expect NYT think-pieces about this and other LLM-created social faux pas before the end of January.
-
Using ChatGPT to insult coworkers makes them ignore you
By
–
This is already happening. Screenshots of bewildered comments on a Google Doc of hallucinated nonsense make for great office gossip. Once you insult someone at work with ChatGPT replies or draft writing, they'll never read another word you say.