Inspired by the heptagon case, I wanted to test how #chatGPT4 is thinking* in terms of geometry.
It seems it takes the filename into account when answering about the structure's shape, but tricking it into calling it a triangle is a bridge too far.
Chestnut flour 300g. A pinch of salt. About 700ml cold water.
Grease a round or rectangular pan with olive oil (30cm diameter). Pour in the quite liquid mix, add pine nuts and raisins, drizzle with a bit of oil (not too much).
Bake at 180°C for 50 mins. Should form a slight crust on top. Best served warm.
I asked #ChatGpt4 if one can trust #UMAP clusters when experimental validation is not possible (that would be the Nature paper #AllofUsResearch under discussion). My non-nuanced answer based on its nuanced answer is NO
Ich habe mal eine neue 1.2 Release der #mastowall zusammengezimmert, die ein paar Optimierungen für sehr dynamische Lagen wie bei #antifasticker enthält:
🎯Automatiser les critères de réussite des apprenants #IA🤖 #chatgpt4. Un article très intéressant de @JnSaintrapt . Une démarche très inspirante 🌟🙏 https://lien.il-di.com/3tK6ScW
Yesterday, Microsoft released Copilot Pro -- and it's much faster than ChatGPT4!
You can also create images -- via DALL-E 3 -- in Copilot Plus, that will take you to the new "Microsoft Designer" for more detailed editing with 100 daily credits.
I always test my favorite prompt "Cat Riding a Motorcycle" to see how these new image Bots behave, and here are the results!
Not terrible! Sort of, almost, anatomically correct.
The natives on the OpenAI discord server are getting restless. More and more people are posting about how terrible ChatGPT4 has become, and are angry over the fact that OpenAI has done nothing to address the issue, and seem intent on hiding the problem from being posted their status page.
ChatGPT4 has added value in simple analysis tasks you can do with other software but simply works easier and more intuitive, for example making this overview of my followers per instance. Top 10 as separate categories, the rest as "other", ofcourse i can do this in excel, but with ChatGPT it was a lot faster, What's interesting is that it shows the fediverse is indeed less centralized, i follow a majority of people from the smaller instances. #chatgpt4#data#analysis#experiment#mastodon
Some free advice, if you are betting your business on quickly "solving the hallucination problem" so you can fully integrate LLMs, then you might want to find out what hallucination really is. Just saw people being optimistic because of this article on the topic:
Just tried using #chatGPT4 to produce a fun christmas 'gift certificate' and it was nearly impossible to get right. The misspellings, the presentation, the orientation were all not only variable, but impossible to tell it "just do what I asked you".
Of course, I may be able to get around this with clever "prompt engineering" but after 4 tries I wasn't anywhere close. #chatGPT
Google announced #Gemini yesterday, a new multimodal #AI. A demo video shows off its ability to process speech and video input, generating both text and speech outputs.
However, the demo is edited, omitting potential mistakes and we don't know what the final UI will be. It appears to surpass #chatGPT4 in some ways, but I would not be surprised if openAI beat them to market with some of the features shown. Watch the demo here:
If you take screen shots from the video and feed them to #chatGPT4, it has little trouble recognizing the same things that #gemini is doing. I would be surprised if openAI's models didn't have similar capabilities. It's mainly a difference of UI.
Notice that it recognizes that it could be a duck one step than #Gemini.
@ErikJonker A year ago, I didn't think #LLMs could do anything interesting. But the leaps from chatGPT3 to 3.5 and #chatGPT4 were sizable. I think we will continue to see incremental improvements for a while, each one not huge, but when we look back the changes will be remarkable.
I wouldn't downplay the importance of a good UI. Google's big contribution may be in how we use AI. Imagine being able to point your camera at anything and say, "What is that?" or having that embedded in glasses?
Die Suche nach der perfekten Kapitalanlage stellt für viele Investierende eine große Herausforderung dar. Prof. Lars Hornuf von der #TUDresden hat untersucht, ob und inwieweit KI-Tools wie #ChatGPT bei der Anlageentscheidung beraten können. Die Ergebnisse zeigen, dass die Portfoliovorschläge der künstlichen Intelligenz mit denen professioneller Anlageberater vergleichbar sind.
Wir verschicken jedes Jahr Weihnachtskarten und dieses Jahr dachte ich: Hey, wofür gibt es KI, die uns die ganze Arbeit wegnehmen wird? Also habe ich eingegeben "4 Person family two children man with beard funny christmas card" und naja, vielleicht dauert das noch bis die KI uns die Jobs wegnimmt.
#chatgpt4 is not super good with #nix ...and it's not that different from what I do, i.e. : try stuff until it works.
The difference is that when I try something I'm not as confident as that thing :D
But the end result is not so bad. I learn enough fixing its errors, and at the same time I rarely get stuck.
I'm getting some real benefits out of it.
It's like peer programming with another noob.
Everyone stop. #OpenAI, stop. I got too many side projects.
Now they want me to check and see if you give #ChatGPT4 Turbo a 300 page book, will it read it and be able to answer based on it, or will it ignore it or will it read it but still answer from it's other knowledge or will it be like prompt text and now that is all it can think about.
AI wird definitiv einige Leute den Job kosten. Aber nicht die Personen, bei denen man das initial dachte. #ChatGPT ist einfach mittlerweile so abartig schlecht.
I have nothing to say that will change the current madness, so in the meantime please enjoy some frivolous whale images made in DALL•E 3. I have created an entire new genre (#whalepunk).