ChatGPT has no idea it's bad at math. It thinks it has a calculator, but it doesn't. Claude, to its credit, will own up to its limitations, but its ability to spot when it shouldn't answer is not reliable:
@goodside
-
Claude AI Gains New Capability: Admitting When It Doesn’t Know
By
–
But for the *really* hard problems, Claude has a new trick — it can admit it doesn't know:
-
Claude’s Calculation Limitations: Hallucinations Compared to ChatGPT and Calculator
By
–
As a counter-balance to all the positivity I'm posting about Claude (a new model from @AnthropicAI
), starting a thread on limitations. First, for many calculation problems, hallucinations are just as bad as in ChatGPT — here's ChatGPT vs. Claude vs. a calculator: -
Generating Comedy with Claude AI is Easy, Demonstrates Endless Creative Potential
By
–
Generating comedy with Claude (new model from @AnthropicAI
) is crazy easy. I’m holding back so I don’t misrepresent it, but this stuff is endless and you can filter/workshop it all day. If you bring it up to speed on “Jenna Ortega Reveals”, these are its first 10 tries: -
ChatGPT vs. Claude: Comparing AI Models with a Literary Prompt
By
–
Side-by-side comparison: @OpenAI
's ChatGPT vs. @AnthropicAI
's Claude Each model is asked to compare itself to the machine from Stanisław Lem's "The Cyberiad" (1965) that can create any object whose name begins with "n": -
Claude AI Model Compared to ChatGPT for Factual Accuracy
By
–
Compared to ChatGPT, @AnthropicAI
's new model, Claude, feels somewhat less likely to make up facts, but the tendency is still there. E.g., when discussing what happened on each season of ABC's Lost, its memory is even more confused than most people's: -
Morty’s take on Anthropic’s Constitutional AI for AI safety
By
–
Or, if you prefer, in the style of Morty: "But I guess Anthropic's researchers are hoping that Constitutional AI and this principles-based approach is the way to go to keep AI systems like me from destroying humanity or something."
-
Claude AI Demonstrates Enhanced Self-Awareness Compared to GPT-3
By
–
Compare to GPT-3, Claude (a new model from @AnthropicAI
) has much more to say for itself. Specifically, it's able to eloquently demonstrate awareness of what it is, who its creators are, and what principles informed its own design: -
Demo of Anthropic’s Claude AI Model for Seinfeld Episode Creation
By
–
A few weeks ago I got early access to a new model from @AnthropicAI
. Its name is Claude. To demo, Claude will give the first scene of a Seinfeld episode where Jerry receives the supposed key to AI alignment from @sama
, ignoring dire warnings from Kramer. The prompt: -
Cognition in LLMs: Associative Memory vs. Functional Awareness
By
–
That’s a good point. What humans see as cognition in LLMs is often just good associative memory, or (generously) intuition. A model that can echo “I am aware” is not necessarily, but in larger LLMs more of the functional implications of awareness begin to apply.