Google’s Bard Is Now Gemini - What’s New?

Even if it can't beat Chat GPT (which I think is likely), the fact that it's more integrated should make it much more useful to the average user.

Tried to use it to translate something from a manga and it said it can't because people are in the image lol

Not very impressed. I tried the image gen feature and since I have a black poodle I wanted to generate an image of a black poodle running through a summer meadow. It refused to generate because my prompt was "inappropriate". I thought... well maybe you aren't allowed to use the word "black" anymore nowadays, so I tried again without it. Apparently that was still too inappropriate for Gemini. Gets the award of most paranoid and restricted AI LLM from me. Congrats google.

"...an assistant that works for you..." - No. It works for Google.

It's perfectly fine if GPT-4 is still better than Gemini. The important thing here is that there is competition, for the consumer benefits from it, and it also pushes advancements faster. With competition, everybody wins.

I asked Gemini about mechanical locks. It claimed that disc detainer locks can be raked open, which is obviously false. I would say it's a bit worse than ChatGPT but it's already much better than Bard. Unlike GPT which can correct itself when you question about the details of incorrect part, Gemini was happy to enter deeper into the hallucination. And I was pleasently surprised that Gemini was happy to discuss about mechanical locks. Bing variant of GPT doesn't like this kind of questions.

I've tried it and i didn't like it, hallucinations are very often, and it refused to do things that it can like generating images or checking mail

I've been comparing Gemini Ultra and GPT-4 the moment it came out. For my use-cases (mainly programming / code review) GPT-4 is slightly better. Gemini does seem more 'creative' in it's answers both in programming and writing. One major deal breaker for me is that Gemini Ultra context size is only 32K tokens compared to the 128K of GPT4-turbo. That means it can't read large files. Another super annoying issue is no file upload?! I can upload images but no CSV, PDF etc. Why? One awesome thing Gemini can do is summarize Youtube videos! Which is incredibly useful. I've let it summarize hours of lectures / tutorials.

Just asked it about some movies and gave me a bunch of nonsense. The "movies" either straight up don't exist, or are completely different than what it describes. I mean, it claimed that "Moonwalker" (1988 Michael Jackson music video/movie) was a Turkish movie about an astronaut stranded on the moon! Google really messed this up to the point of being completely unusable.

can you make one for the new Gemini 1.5? I love your content, it gives me a great feeling. keep it up 👍

very exciting. you are the best.

I tried it, but now it still looks like an idot, not like OPENAI.

Cheers to this new era 🎉

Quite the energetic synthetic voice I love it!😁❤👍😊

I only use the google assistant to set alarms... and it sometimes fails at that.

The one thing it seems to excel at is creative writing. Outside of that I find Prob and Ultra lacking. With the speed of responses, 3 drafts, no limits, and the fact that it's good at creative writing I would say it's worth using at least the free version for some things, but the Ultra model may also be good for some Python programming. I'm just glad we're getting competition.

I'd be curious to see a comparison in terms of efficiency. As in, how many trees it has to burn to get to the same result. :)

LLMs are too useful to ignore, yet so frustrating to talk to. Bizarre moral hang-ups, dementia-like forgetfulness, paranoia, and over-explaining basic concepts.

Can i know why it always give me some random lyrics when i ask to tell me lyric of any song?

It is not as good as GPT-4 that's for sure. However, it is faster, has no cap, and has more personality(a little bit more like pi assistant). The biggest issue really is that Google hyped this to the moon, and it didn't deliver at all on the promises. If they said that it is a competitive model rather than the best model ever made, that would have led to much less criticism and disappointment. I remember Demis Hasabis claiming that it would "eclipse GPT-4". When saying stuff like that, people are not going to listen to you in the future.