My corporate workspace which seem to be a pro account is limited to 3.6 but my personal account at 20/month has given me the latest model always immediately. Flash 3.8 is my go to model for pretty much everything those days
Very often a demerge comes before an acquisition. Example, a fund could buy a huge group/conglomerate made of various companies, and then sell each companies separately. Each company sold is merged to another group but demerged from its original group.
same experience. I was disappointed with 3.6 flash, but 3.7 seems a game changer IMO. Gemini 3.7 flash is really really near frontier intelligence at an order of magnitude cheaper and an order of magnitude faster. The speed is a major feature, I feel this is overlooked
This is a refreshing opinion, and also matches with my own experience. Many software engineers around me have now made the conclusion that AI will take their jobs and they're thinking changing careers already. I feel this is too early to tell. The prompts I write are all very technical, someone without my expertise would struggle with just an agent to talk to. Every time I do something outside my expertise, it is not as fast as one would imagine. Expertise do help tremendously and keep things in order
Frenchman here, living in Spain. This speaks to me on many levels. Bizum is so integrated in my daily life in Spain, that I wished my french friends had it when we need to transfer money between each other. Looks like we're going in that direction. Phenomenal
Most of the successes, especially online, is rarely about the thing that is built but more about the marketing around it. I don't we can fully automate marketing effectively
This is colossal. It can creates embeddings on pretty much any type of format, video, audio, documents. The context is still a bit small compared to what we are used to in text, but this seems major
I am no expert on the matter but I always thought ternery weight should be part of the neural net nature, trained on those, rather than a compression mesure for inference. Are they any training made on ternery weight models that are proven to be effective?
I switched a very long time ago when Gemini was released and it was a very easy switch at the time. I have never missed ChatGPT and due to current circumstances I'm kind of happy I made the switch. It woukd be a lot harder for me now to switch from Gemini (except for code of course)
You may pay to ChatGPT, but sooner or later you will become their product too. All the conversations you had or will have will be turned into signals to match you with products from advertisers, maybe not directly in the conversation with them, but anywhere else. It's not a mater of if, but looking at the pace things are going, and how financially pressured openai is, it's only a matter of time that their conversations with them will be turned into profit in some way or another, they basically have no choice financially.
Agreed, also worth pointing out that Google still owns 14% of Anthropic + Anthropic is signing billion dollar scale deals with Google Cloud to train their models on their TPUs. So Claude success indirectly contributes to Google success. The AI race is not only about the frontier models.
My experience with Antigravity is the opposite. It's the first time in over 10 years that an IDE has managed to take me out a bit out of the jetbrain suite. I did not think that was something possible as I am a hardcore jetbrain user/lover.
I'll agree to disagree. In any thread about a new model, I personally expect the pelican comment to be out there. It's informative, ritualistic and frankly fun. Your comment however, is a little harsh. Why mad?
Depending on how you look at it I suppose but I believe Gemini surpasses OpenAI on many levels now. Better photo and video models. The leaderboard for text and embeddings are also putting Google on top of Openai.
A junior in SQL would need AI to write things they're not sure about, the same way stackoverflow has helped us for many many years before AI. A senior in sql, and in fact any languages, would use AI to be accelerated (I know I do).
I second trafilatura greatly. This will save a huge amount of money to just send the text to the LLM.
I used it on this recent project (shameless plug): https://github.com/philippe2803/contentmap. It's a simple python library that creates a vector store for any website, using a domain XML sitemap as a starting point. The challenge was that each domain has its own HTML structure, and to create a vector store, we need the actual content, removing HTML tags, etc. Trafilatura basically does that for any url, in just a few lines of code.
Actually, Sqlite-vss has been untouched for quite some time, and the creator has officially communicated that it was deprecated to be replaced by sqlite-vec, which has recently seen its first non-alpha release (v0.1.0). So, I would embrace sqlite-vec now if I were you.
I have not used sqlite-vec much because it was only alpha-released for now, but it finally came out a few days ago. I'm looking into integrating it and use it to make sqlite more my go-to RAG database.
Just submit your XML sitemap into a python class, and it will do the crawling, chunking, vectorizing and storage in an SQLite file for you. It's using SQLiteVSS integration with Langchain, but thinking of moving away from it, and do an integration with the new sqlite-vec instead.
Very happy to see this extension already out. I tried some of the previous alpha version and is incredibly much easier to use and integrate than the previous sqlite-vss extension. Kudos to the creator.
I originally added sqlite-vss (your original vector search implementation) on Langchain as a vectorstore. Do you think this one is mature enough to add on Langchain, or should I wait a bit?
Love your work by the way, I have been using sqlite-vss on a few projects already.
Author of the article here. Just went though your website and I can not believe I never heard about Mojeek. I'll probably have a go at your API eventually.