Free Gemini gets better and faster, European users will get it in Google Messages

26 July 2024
Google infuses the free-tier Gemini with 1.5 Flash with improvements in quality and latency.

Sort by:

jake, 27 Jul 2024Unfortunately it is far behind GPT4. FAR behind for everyt... moreIt's the problem of internet enabled LLM, the search result can actually poison the model if the priority is borked, e.g. Gemini knows smoking is bad but if it searched Google whether smoking is bad or not and the results from the web are wrong, saying smoking is good then Gemini will parrot and justify it.
Ironic that it's one of the method we use to reduce hallucination but it's actually worse in some way.

    jake, 27 Jul 2024Unfortunately it is far behind GPT4. FAR behind for everyt... moreWhile I fully agree ..

    It's the integration within Android, especially Samsug phones that's the real take home use case here for the end user.

    Something the EU is investigating to make a case out of to swe if it prevents AI competition to be used or their performance vs the competition. It's possible this nay be the case.

    Then again where was the EU when it comes to:
    Siri or Google Assistant?! Where wrre they when Google purchased HTC fully absorbed to directly compete with theor partners whom are also string-armed customers to their os and software ??!!

    This seems like a knee-jerk reaction to OpenAI Search right now in minutes beta

      • j
      • jake
      • LTH
      • 27 Jul 2024

      Anonymous, 26 Jul 2024But is it better than GPT 4o mini in at least one area? If ... moreUnfortunately it is far behind GPT4. FAR behind for everything from simple question to complex queries. It's like comparing Yahoo vs Google for search results... or ferari vs honda

        • ?
        • Anonymous
        • 3SI
        • 26 Jul 2024

        But is it better than GPT 4o mini in at least one area? If no, I see no reason to use it.

          • ?
          • Anonymous
          • Bnx
          • 26 Jul 2024

          At least the previous version was absolutely useless garbage.

            Another garbage to debloat & uninstall

              Anonymous, 26 Jul 2024"ChatGPT free tier has GPT-4o-2024-05-13, arguably the... more>Best tested LLM is Claude 3.5 Sonnet, it wins in 90%+ benchmarks.
              That's synthetic benchmark, it is known that dataset can be tainted or tampered to make a model perform better on that specific test set while not on the other. e.g. model trained in GSM8K test set will have 100% in that but not MATH. Everyone worth their money knows it's unreliable at best, look up Open LLM Leaderboard to find out why it's bad. I'll tell you what, my finetuned Mistral-7b beat GPT-4 back in the day (lol). I'm using human preference benchmark, basically human test the model and vote which one is best.
              >LLaMA-3.1-405b-instruct weights 810 GB in FP16 and 405GB in lower precision FP8. Which graphics card for PC has at least 405GB memory?
              Well duh, I did say if you have the compute, which is applicable for HPC with at least 256GB RAM for minifloat, contrary to popular belief, you don't need CUDA to run AI application, not even a GPU is needed if you're desperate. The point is now every company can finetune L3.1 to potentially give OpenAI a run for their money, you can see their damage control by suddenly offering free finetune for GPT-4o-mini several hours after L3.1 release.

                notafanboy, 26 Jul 2024Did you use AI to generate this reply?I'm using the unreleased Q*.

                  LOVE PENGUIN, 26 Jul 2024ChatGPT free tier has GPT-4o-2024-05-13, arguably the best ... moreDid you use AI to generate this reply?

                    • ?
                    • Anonymous
                    • mNr
                    • 26 Jul 2024

                    "ChatGPT free tier has GPT-4o-2024-05-13, arguably the best tested LLM"
                    Best tested LLM is Claude 3.5 Sonnet, it wins in 90%+ benchmarks.
                    Other thing - ChatGPT web is NOT comparable API's version, as web version has system prompt that make it more safe, so it uses less knowledge that GPT-4o-2024-05-13 has.

                    "LLaMA-3.1-405b-instruct also looks promising and it's free for all, you can run it on your PC if you have the compute"
                    LLaMA-3.1-405b-instruct weights 810 GB in FP16 and 405GB in lower precision FP8. Which graphics card for PC has at least 405GB memory?
                    I thought you need Nvidia Hopper Cluster to run LLama 405b, so please enlighten me which PC can handle it. By PC you mean ones with 17x RTX 4090 GPU (408GB total)?
                    Or are you talking about some heavily quantized version which still will require you 4x RTX 4090, but deliver results that aren't better than free web versions of ChatGPT, Claude or Gemini?
                    LLama3 405b is great model that in FP8 fights with GPT4o and wins against Gemini, but saying "just run it on PC" is... just not true. If it would be true then OpenAI would be completely dead.

                    I'm not even calculating overhead of CUDA, context lenght (easily additional 100GB+) or using full precision FP16.

                    By the way guys - Gemini 1.5 Flash and a lot better Gemini 1.5 Pro are both FREE (and were free from many months), just use AI Studio instead of Gemini app. Gemini 1.5 Flash is not great, but 1.5 Pro is very good.

                      justasmile, 26 Jul 2024Explain how?ChatGPT free tier has GPT-4o-2024-05-13, arguably the best tested LLM in the market right now while Gemini-Advanced-0514 that is exclusive to paid tier is ranked 3rd-4th. LLaMA-3.1-405b-instruct also looks promising and it's free for all, you can run it on your PC if you have the compute, synthetic benchmark shows it beat 4o in some key points but human preference testing for all L3.1 family is still ongoing. As for this model in the news, Gemini-1.5-Flash-API-0514 is 13th currently.

                        TUSHAR , 26 Jul 2024I use paid gemini, and from my experience, even free chat g... moreExplain how?

                          Google is fighting. Be careful ChatGPT

                            • T
                            • TUSHAR
                            • CbE
                            • 26 Jul 2024

                            I use paid gemini, and from my experience, even free chat gpt is better than gemini advance

                              1 of 1