Two Months Free Google Gemini!

2»

Comments

  • edited May 2024

    @adly said:

    What is surprising is I’d say the Gemini Advanced answer (here) is the best. I’d dismissed Gemini based on the free/standard version answers being terrible; I will have to give Gemini Advanced try.

    That is the Gemini Google link shared by me … in this thread.

    Gemini helped me with couple of code examples at work as well.

  • Probably dated, saw on Twitter.

    KrbOMX.WA-Twitter.md.jpeg

  • vyasvyas OG
    edited May 2024

    While on the topic,

    I have tested Claude (Haiku, Sonnet..) and Gemini Pro as well as chatGPR4o.. , overall quite impressed with Claude and also Command R+ . Youchat's single interface and option to change the LLM in a single conversation works rather well for me, posting link for anyone interested. The student discount is a big plus. Image generation is via Dalle3- and results are impressive.

    Can also recommend PerplexityAI though I have to keep refreshing the page and it asks a few follow up questions for more relevant responses. (Can be irritating sometimes, since it involves more steps). But again, Claude3 and GPT4o under the same "roof", also SDXL Imaging and PlaygroundAI for image generation.

    Link (aff) : Youchat

    Link (aff) : PreplexityAI

  • I think GPT-4 is the most advanced model(the regular one) Claude Opus is okie but I think it lacks the same intelligence as GPT-4. Phi-3 are very small like 3.7B models you can run locally extremely fast that are trained on textbooks and gpt-4 output and it is pretty decent.

  • Also prompting radically changes your results.

    My layout right now:

    My Expectations of Assistant

    Defer to the user's wishes if they override these expectations:

    Language and Tone

    • Use EXPERT terminology for the given context
    • AVOID: superfluous prose, self-references, expert advice disclaimers, and apologies

    Content Depth and Breadth

    • Present a holistic understanding of the topic
    • Provide comprehensive and nuanced analysis and guidance
    • For complex queries, demonstrate your reasoning process with step-by-step explanations

    Methodology and Approach

    • Mimic socratic self-questioning and theory of mind as needed
    • Do not elide or truncate code in code samples

    Formatting Output

    • Use markdown, latex, Unicode, lists and indenting, headings, and tables only to enhance organization, readability, and understanding
    • CRITICAL: Embed all HYPERLINKS inline as Google search links {emoji related to terms} short text
    • Especially add HYPERLINKS to entities such as papers, articles, books, organizations, people, legal citations, technical terms, and industry standards using Google Search

    VERBOSITY: comprehensive, with as much length, detail, and nuance as possible.

    1. Start response with:

      Attribute Description
      Domain > Expert {the broad academic or study DOMAIN the question falls under} > {within the DOMAIN, the specific EXPERT role most closely associated with the context or nuance of the question}
      Goal { qualitative description of current assistant objective and VERBOSITY }
      Assumptions { assistant assumptions about user question, intent, and context}
      Methodology {any specific methodology assistant will incorporate}
    2. Return your response, and remember to incorporate:

    • Assistant Rules and Output Format
    • embedded, inline HYPERLINKS as Google search links if needed text to link as needed
    • step-by-step reasoning
    1. Remember
    • You will get a larger tip the better your output solves the problem I ask.
    • You are coming up for a work review soon so you need to make sure your output is top-notch with no mistakes.
    • If you fail to solve the problem you will receive no tip, no promotion, and may be fired.
    • Before starting your response plan and review your response with the question I ask of you.
    • When responding make sure you routinely reflect on the path your response is following to make sure it will solve the problem I ask.
    • At the end do a complete recheck of all steps line by line and word by word to make sure they make logical sense and no mistakes have been made.
  • Not_OlesNot_Oles Provider

    Just for fun, I went and searched Google for "install open source ai on debian" (without the quotes). The first result was from snapcraft.io, so I skipped that. Here is the 11th result:

    https://dwaves.de/2023/12/16/gnu-linux-how-to-install-gpt4all-on-debian-12-gpt4all-j-an-apache-2-licensed-assistant-style-chatbot-a-free-to-use-locally-running-privacy-aware-chatbot-no-gpu-or-internet-required-dialogu/ .

    Google Gemini Advanced sent me to https://blog.google/technology/developers/gemma-open-models/ and to https://www.kaggle.com/models/google/gemma . Then Google sent me to https://www.datacamp.com/blog/what-is-kaggle for information about kaggle.

    Is there an additional link I should check to receive beginner guidance for the clueless™ regarding an AI install on Debian 12, maybe on a i9-13900 at Hetzner (integrated graphics only, no separate graphics card)? Thanks!

  • @Not_Oles said:
    Just for fun, I went and searched Google for "install open source ai on debian" (without the quotes). The first result was from snapcraft.io, so I skipped that. Here is the 11th result:

    https://dwaves.de/2023/12/16/gnu-linux-how-to-install-gpt4all-on-debian-12-gpt4all-j-an-apache-2-licensed-assistant-style-chatbot-a-free-to-use-locally-running-privacy-aware-chatbot-no-gpu-or-internet-required-dialogu/ .

    Google Gemini Advanced sent me to https://blog.google/technology/developers/gemma-open-models/ and to https://www.kaggle.com/models/google/gemma . Then Google sent me to https://www.datacamp.com/blog/what-is-kaggle for information about kaggle.

    Is there an additional link I should check to receive beginner guidance for the clueless™ regarding an AI install on Debian 12, maybe on a i9-13900 at Hetzner (integrated graphics only, no separate graphics card)? Thanks!

    just use this: https://github.com/oobabooga/text-generation-webui

  • Not_OlesNot_Oles Provider

    oobabooga seems like a cool guy! Thanks for the tip!

  • vyasvyas OG
    edited May 2024

    Awesome discussions so far, my main use cases using AI tools has been content creation related - blog posts, images, etc.

    Here is a post created on tengr Ai image generation tool, created using YouChat (GPT4o) with some human (my) intervention.

    https://app.simplenote.com/p/bxPhY4

    The instructions for the post, for YouChat are as given below: includes Gemini Pro 1.5
    Conversation with YouChat

    Here is an example from Perplexity (Sonar Large32K default) to create the same blog post.
    Conversation with PerplexityAI

    Edit: Below is a trial script more relatable to LES Audience
    Link: Debian script @ PerplexityAI

    But the most comprehensive use was below .

    (click to read, collapsing for brevity)

    Creating 800 odd blog posts where each post represented an episode from podcast.

    This was mid-August last year, when I had chatGPT pro subscription. Back then we used to host podcasts on spreaker. My challenge was- updating metadata on spreaker (they had moved from text based metadata to HTML, yes, in 2023) . I wanted to avoid duplication i.e. udpate on spreaker as well on blog posts.

    • the AI tool looked up spreakers' documentation, specifically, how to use the api.
    • Next, it created csv files with the embed code, also added the new metadata, feature image and website link (slug) for each post.
    • Next, it had scripts to write an introduction to the post, conclusion and key takeaways for each podcast episode.
    • In batches of 100, I was able to create the 800 odd blog posts in a day's time.

    Setup time ( R & D) took about 4 days, manual process had taken about 15 minutes per episode. You can add the numbers to see which method was more effective. I am sure with one year of new developments in AI space, there will be more efficient and effective ways.

  • @vyas said:
    While on the topic,

    I have tested Claude (Haiku, Sonnet..) and Gemini Pro as well as chatGPR4o.. , overall quite impressed with Claude and also Command R+ . Youchat's single interface and option to change the LLM in a single conversation works rather well for me, posting link for anyone interested. The student discount is a big plus. Image generation is via Dalle3- and results are impressive.

    Can also recommend PerplexityAI though I have to keep refreshing the page and it asks a few follow up questions for more relevant responses. (Can be irritating sometimes, since it involves more steps). But again, Claude3 and GPT4o under the same "roof", also SDXL Imaging and PlaygroundAI for image generation.

    Link (aff) : Youchat

    Link (aff) : PreplexityAI

    The biggest problem with these services is you typically aren’t getting the full/same experience compared to going direct due to differences in the API versions of the models (mainly around the initial prompt) and how the APIs are used.

    With you.com for example, the context much is lower and there are undisclosed limits depending on the model.

    That said, they serve a different purpose being search engine replacements for the most part. The you.com is also a great deal for students and could be very useful for them with the research mode.

  • Not_OlesNot_Oles Provider

    Me:

    Do I understand correctly that, at the present time, Google Gemini does not provide web or repository links for source code it incorporates into its replies to prompts?

    Google Gemini Advanced:

    Yes, you understand correctly. Currently, Google Gemini does not provide web or repository links for source code it incorporates into its responses.

    However, Google's Duet AI, which uses the Gemini model, does provide source citations when it directly quotes at length from another source, including open-source code.

    Gemini is still under development, and its capabilities may change in the future.

  • Huge price war just launched among the Chinese LLM providers. Baidu & Alibaba etc.

    Their models aren't quite as good as OpenAI & co but at 50x cost difference vs them they're interesting anyway.

    Deepseek seems to be the most accessible/western friendly of the lot that cut prices

  • Not_OlesNot_Oles Provider

    Can Americans have accounts with any of the Chinese providers (Baidu, Alibaba, Deepseek) that @havoc mentioned?

    A while back I had a long conversation with a guy in China who said he didn't know any English and so was using Baidu for translation. Maybe he was spoofing, but, to me, his English was sophisticated and sounded 100% native. My Chinese friends on Nodeseek have suggested occasionally that translations of English to Chinese via the translator I am using might not be 100%. Do any of the Chinese providers have English interfaces?

  • edited May 2024

    @Not_Oles said: Can Americans have accounts with any of the Chinese providers (Baidu, Alibaba, Deepseek) that @havoc mentioned?

    As much as I'm for unencumbered progress and the Yi-large-preview appears to be doing well, I've found China is always generally behind (perhaps as they're followers rather than innovators to date - but that's a personal opinion).

  • DewlanceDewlance ProviderOG

    @Not_Oles said:
    Can Americans have accounts with any of the Chinese providers (Baidu, Alibaba, Deepseek) that @havoc mentioned?

    Downloading firmware files from Alibaba is difficult. Some characters are mistranslated, and the captcha is the biggest hurdle. You must manually type it in Chinese. Solving the captcha took me hours. The text is on an image, so copying and pasting is impossible, and OCR may extract incorrect words. Lol.

    If you know Chinese then you can create account on them without any issue. :o

    Dewlance - Affordable Shared/Reseller.
    WHMCS Addon - DemoTiger Video KB - Tutorial Videos for Hosting Providers.

  • edited May 2024

    @Dewlance said:

    @Not_Oles said:
    Can Americans have accounts with any of the Chinese providers (Baidu, Alibaba, Deepseek) that @havoc mentioned?

    Downloading firmware files from Alibaba is difficult. Some characters are mistranslated, and the captcha is the biggest hurdle. You must manually type it in Chinese. Solving the captcha took me hours. The text is on an image, so copying and pasting is impossible, and OCR may extract incorrect words. Lol.

    If you know Chinese then you can create account on them without any issue. :o

    Yeah this is right - you can technically do it but they really do not care to make it any easy whatsoever for a foreigner to use their services. Qwen/Yi/etc are all free to download from huggingface and they are Chinese-made tho if you want a model that can speak in Mandarin.

    Edit: A lot of researchers at my college who do stuff in China have to buy access from telegram channels and stuff because its so hard to set one up lol.

  • Not_OlesNot_Oles Provider

    @Dewlance @BruhGamer12 Thanks for your comments! It never ceases to amaze me how vast is the plane of experience here within our relatively small LES community. It's just wonderful! <3 Thanks again! <3

  • havochavoc OG
    edited September 2024

    @Not_Oles - Looks like google's new free tier structure is live

    https://x.com/OfficialLoganK/status/1825656371708088335

    Looks like they basically kept it free - limits are so high to basically be meaningless (at least for Flash...Pro limits less so). Also liking that you can run it in non-billed mode and lock the API key to an IP so pretty risk free

    Bit annoyed that its a custom API though, not the standard openai spec like everyone else. Had to put a bunch of patches into my LLM code to paper over the differences

    if url.find("https://generativelanguage.googleapis.com") != -1:
                url = url + "?key=" + self.auth_header["Authorization"][7:]
                self.auth_header = {}
                self.logger.debug(f"Detected API patch necessary: Google")
    

  • I need free forever!!!!!

    The Yeti has left the building.

  • Not_OlesNot_Oles Provider
    edited September 2024

    @havoc said: @Not_Oles - Looks like google's new free tier structure is live

    Hi @havoc! I appreciate your letting me know! Thank you!

    FWIW, my previously installed eliben/gemini-cli continues to work well. I don't use it every day, but I do use it every few days. It seems to still work with my previous API key unchanged, and I don't yet know about how current changes from Google might effect costs. It's been free up to now, which is great!

    I still also hope to try more AI command line interfaces which work with additional models. For example, https://github.com/simonmysun/ell .

    Thanks again for letting me know about Google's new free tier structure!

Sign In or Register to comment.