Listening to very smart people talk about #GPT4 I'm reminded of the joke about a checkers-playing dog.
A guy has a dog that plays checkers. "My goodness," everyone says, "that's amazing. What a brilliant dog!"
"Not really," he replies, "I beat him four games out of five."
That's GPT4. It's capacities are amazing and completely unexpected.
But it's also so limited. You shouldn't back the dog in a checkers tournament, and you shouldn't use an LLM as a medical assistant or in many other ways.
After hearing Sebastian Bubeck talk about the #SparksOfAGI paper today, I decided to give #GPT4 another chance.
If it can really reason, it should be able to solve very simple logic puzzles. So I made one up. Sebastian stressed the importance of asking the question right, so I stressed that this is a logic puzzle and didn't add anything confusing about knights and knaves.
ICYMI: Do you use #ChatGPT? Want to help a #PhD candidate with their #research? My colleague at #ANU's School of #Cybernetics, @nedcpr, is currently running a #survey - and he would love your help by filling it out:
Imagine you're blind, and you're excited because this very thought-provoking article you're reading is about to break down a bunch of covid data to you in a chart. You then get down to the alt text for said article, and all the alt text says is 'Chart showing the numbers of covid cases in each state from the years 2020-2023...' The article then presumes to assume that you can read all the data points on the chart. You'd be thinking, 'Well thank you for providing alt text, but I still have no data points.' You'd be frustrated because the article is referencing data points to which you have no access. You might even think, 'Well, I appreciate the conclusions brought out in this article, but without accessing the chart, I can't fact check the statistics mentioned by the article's author.'
Moral of the story? Charts absolutely deserve good alt text.
Pascal Laliberté, I'd like to have text descriptions above charts. Before today, generated descriptions were tough.
Charts deserve good alt text!
Soon I think this will be everywhere with the help of LLMs. For now, it's slow-ish, costly-ish, but possible with OpenAI's gpt-4 model.
My little experiment got me this description for the chart below:
"Team met 4.5-hour first response goal on average. Maximums occasionally exceeded goal, with highest at 6.57 hours."
Not bad
#llms#ai#openai#gpt4#accessibility Media description: Screenshot of a chart showing the time to first response to support cases, the min/max range and average daily, plus a chart for support case volume below.
3 hours ago. Public.
On that CNET thing in the last boost, my first thought was "this is gonna make search even more useless" and… yeeeep "They are clearly optimized to take advantage of Google’s search algorithms, and to end up at the top of peoples’ results pages"
#OpenAI admits #GPT4 "still makes many of the errors of previous versions, including 'hallucinating' nonsense, perpetuating social biases and offering bad advice" but implies (largely unchallenged in the press) that these are merely bugs to be fixed, not fundamental characteristics of the technology