Personally, the AI studio with vision has been awesome for me. It can see your screen and has helped me learn new platforms. Very, underrated.
They NEED to fix the cutoff issue, its actually one of the better canvas/artifacts offering outside of that issue
Canvas is still a bit half-baked, but like you, I've been very impressed with Google AI inprovements lately. I'm daily diving Gemini now.
The canvas feature is awesome. I had fun making a document about the Qing dynasty. Expanding paras and inserting new details and pics in between, made it an immersive self-learning experience. Tools like these are going to make learning fun and interactive.
Nice content, I think limitations are because output is limited to 8k token , the thinking model has much more output token
The biggest bottleneck with using LLMs is the method we use to create prompts. I think Canvas editors like this are really good, and I hope to see them be more dynamic. Voice input is also pretty good too.
Google did breakthrough in GenAI, images looks very good now. Comparing to what it was before 😄
I've fallen in love with Gemini 2.0 Pro Experimental 02-05 (Free) and Gemini 2.0 Flash Thinking Experimental 01-21 (Free).
now perhaps they can fix ‘your gem?’
That's because they're stealing all my data all my work
Dude you seem like you might be open to a weird idea It's not perfect and I'm happy to show you some of the prompt schema I've been working on but if you ask the model to construct an ontology first, not just write me this code, but before you have the right to speak you must construct the framework that the request makes sense in, then generate code. The outputs are way more consistent and way less hallucination prone
I too think google/gemini models are underrated, and OpenAI overrated. I use Claude 70%, Gemini 30%. But for things to summarize, extract, retrieve, - I don't even bother to parse manually any more, I ask gemini-2.0 flash for structured output to pydantic and it never fails and it doesn't really slow things down.
I have been a developer for 20 years and still have years to go before I retire. I have been so enthusiastic about ai, but now Iam just nudging what ever code I need to write. Now Iam in a place where it feels pointless to code, it’s so easy an effortless to achieve what ever is asked for. I can’t be the only one feeling like this. I see it it in my colleagues as well. They used to be proud of their work, and now they are just surviving. We are just sitting around waiting for the models to be smarter, and our contribution less important. Perhaps what Iam addressing is a kind of grief - not just for potential economic changes, but for the loss of a particular way of building meaning through work. Creating something through your own knowledge and effort provides a different kind of satisfaction than orchestrating automated processes, even if the end result looks similar.
Sometimes I mock Google for not creating the best AI model, but ironically Gemini 2.0 is my most used AI.... 😅
$400 a month for gemini advanced! As a hobbyist there is no way.
I’d kind of written Google off as well, but they’re making big strides these days. On the tweet-writing, the UI is very cool and intuitive, but the writing still sounds like an AI, even (or especially) in super-casual mode 😕 Writing like a human is yet to be figured out, but I’m sure coming soon. (6-12 months for indistinguishable)
Gemini is the worst of the major LLMs. But its still good enough to be kindof useful, sometimes. I just feel like there are better models for each specific thing that you might want to do. But if you want a model that does most things at a C+ level, then Gemini is fine.
Once Pro is fully released, I think they will blow everyone else out of the water!
I recently use Gemini image editor to generate background for my lyrics videos, and it is very good
@zstrizzel