Dhruv Sharma, Ph.D.’s Post

Plenty of AI tools claim to be the fastest, but what good is speed if you have to sacrifice accuracy? At ScreensAI we put accuracy at the forefront of everything we do: from the features we release to the way we engineer our platform. That's why we have a 97.5% accuracy rate on a sample of more than 700 human-verified data points. But don't take our word for it, come see for yourself how we're combining your 🧠 expertise - and best-in-class 🎯 accuracy - with AI 🏎️ speed:

This article came across my stream on Mastadon via Baldur Bjarnason. It describes services offering "large-scale sentiment analysis of LMM chatbot responses to brand terms along with an analysis of how keywords, queries, and prompts affect the result." The article calls this generative AI's Altavista moment. Here's my favorite quote: "To understand what’s happening here you need to remember that *it’s a category error to treat LLMs as thinking entities*. They are statistical models that work with numbers – tokens – that represent language and the relationships between the words. It’s statistics about language wrapped up in an anthropomorphic simulation. *It’s not people*." Sound familiar Tony Foster?? 😉 But that's the thrust of the article - if you want to manipulate LLM's you can't treat the chatbot like a human and throw words at it. After all, LLMs are big statistical problems. The article is full of really interesting links, it's a great read! #LLM #generativeAI #AImanipulation #SEO #STS #LLMO https://lnkd.in/g5D4NJ3y

🔥 Taming the Chaos of LLM Outputs with Ollama’s Structured Outputs 🛠️ If you’ve ever worked with LLM models, you know the pain of dealing with inconsistent, messy outputs. Parsing through unstructured text or unpredictable data is a time-sink nobody enjoys. That’s why I’m so excited about Ollama’s Structured Outputs it’s been a game-changer for making AI responses reliable and ready-to-use. I recently wrote about how this feature lets you define a JSON schema to structure AI outputs exactly how you need them. Whether it’s parsing data, analyzing images, or building APIs, this is the tool I didn’t know I needed. In the blog, I share: How Ollama’s Structured Outputs work Real-world use cases (and trust me, there are a lot!) An example from a Car Management System I built that extracts details from descriptions and images. check it out: https://lnkd.in/e6ZT6vaj #AI #Ollama #JSON #LLM

Agentic RAG incorporates AI agents into the RAG pipeline to orchestrate its components and perform additional actions beyond simple information retrieval and generation to overcome the limitations of the non-agentic pipeline. The RAG agent can then reason and act over the following example retrieval scenarios: ~Decide whether to retrieve information or not ~Decide which tool to use to retrieve relevant information ~Formulate the query itself ~Evaluate the retrieved context and decide whether it needs to re-retrieve. https://lnkd.in/dhknpG7C

Sharing an insightful article on building a multi-agentic RAG system with Hugging Face Code Agents and Qwen2.5–7B-Instruct! 🌟 This piece explores how open-source LLMs can power advanced problem-solving and efficient information retrieval with consumer-grade hardware. A must-read for anyone interested in cutting-edge AI applications. 💡 Check it out here: [https://lnkd.in/e2E-qw_u] #AI #MultiAgentSystems #RAG #OpenSourceAI #MachineLearning"

Negative Keyword Rules? Who? Who has the best rules/guidelines you've seen for finding new negative keywords? I'm building a new script to find & suggest new negs, then expand on those with AI. Part rules, part semantics. I'm also pondering the idea of a little competition baked into the tool... to find the rules that work best. But for that I need written SOPs... so who have you seen that does this really well? Ed Leake? Navah Hopkins? John Moran? Susan Wenograd? Jyll Saskin Gales? Brett Curry? Adriaan Dekker? Miles McNair? Bob Meijer? Who should be on the short list for this? Who wants to play?

What I learned Creating 475,000 lines of Code With AI It's no secret I'm a big critic of AI. The Acolytes, like true Acolytes, only extoll the virtues without any connection to actual value. Of course "actual value" may vary... My unvarnished opinion after spending 12+ hours a day, 7 days a week working to an actual commercial purpose? AI is pretty useless. The problem is with the engine itself. At the core it is merely a pattern matching engine hard-coded to love the sound of it's own voice. This causes MAJOR problems for us "regular people". But there is a fix and fortunately it's an easy one. Think about how you have a conversation with another person. There is back-and-forth, rewind, fast forward, 'rabbit trails', 'getting back on track', etc. With an LLM it is a strict linear conversation. And, worse, the LLM can't 'ignore' anything. A single word can "pollute" the LLM's decision space causing it to go off in crazy directions. When you have a "conversation" all of the prompts and all of the responses are sent with the new stuff you enter. This is "context" but also pretty dumb from a practical standpoint. That's because in a normal conversation your partner(s) have the ability to 'ignore', 'forget' and 'go back' - unlike the LLM. Enter the fix. Instead of repeatedly entering new prompts, stay with one prompt and develop it. When the LLM gives you output and you think "oh, that's good" don't rush to enter a new prompt. Go back to the prompt you entered and modify that with the 'good stuff'. Repeat that process until the idea or concept is more fully developed. This is how humans "think" and "communicate" - we often don't know what we want to say until we've "talked it out". If you "talk it out" with the LLM it won't forget the crazy stuff you said to get to your actual point. It thinks that crazy stuff is needed and helpful. But for YOU it's just throw-away. So, throw it away by using a single prompt and editing it with the new, better stuff you and the LLM come up with. This is why LLMs suck at decoding human communication. And since the programmers are probably not the most socially adept people in the world... that's not a surprise. GIGO

Safeguard Your LLM Chatbot With Llama Guard 2 How to apply content moderation to your LLM’s inputs and outputs for a more responsible AI systemContinue reading on Towards Data Science » 𝗪𝗵𝗮𝘁'𝘀 𝘁𝗵𝗲 𝗻𝗲𝘅𝘁 𝘁𝗼𝗽𝗶𝗰 𝘆𝗼𝘂'𝗱 𝗹𝗶𝗸𝗲 𝘂𝘀 𝘁𝗼 𝗰𝗼𝘃𝗲𝗿? 𝗖𝗼𝗺𝗺𝗲𝗻𝘁!

Would you like a ReAct agent that fully takes advantage of the tool-calling functionality of the more recent LLMs such as GPT-4o or Claude 3 (and no Langchain doesn't have one ;) )? Check out this post on how we built one - and check out the (open source) agent itself in motleycrew.ai! https://lnkd.in/df8PXhMz

Safeguard Your LLM Chatbot With Llama Guard 2 How to apply content moderation to your LLM’s inputs and outputs for a more responsible AI systemContinue reading on Towards Data Science » 𝗪𝗵𝗮𝘁'𝘀 𝘁𝗵𝗲 𝗻𝗲𝘅𝘁 𝘁𝗼𝗽𝗶𝗰 𝘆𝗼𝘂'𝗱 𝗹𝗶𝗸𝗲 𝘂𝘀 𝘁𝗼 𝗰𝗼𝘃𝗲𝗿? 𝗖𝗼𝗺𝗺𝗲𝗻𝘁!