Gianluca Venere’s Post

Google's #Gemini Flash is preparing to set to cost of the input tokens at $0.05 per 1 million tokens, a significant drop from OpenAI's GPT-3.5 pricing of $0.06 per 1,000 tokens just two years ago. This drop in cost, along with better capabilities, follows Moore's Law in AI. It shows a 100 times increase in cost-effectiveness and a 10 times boost in intelligence, enhancing competitiveness among providers. Key implications include: 🔹 Wider Access: More affordable AI tools for businesses, researchers, and individuals. 🔹 Increased Innovation: Lower costs encourage more experimentation and new applications. 🔹 Economic Impact: Major shifts in industries reliant on knowledge work. 🔹 Ethical Challenges: The need for responsible use and regulation grows. 🔗 Source https://lnkd.in/d7fWgXFk #llm #ai #costefficiency #genai

Structured Outputs, a game changer for developers? OpenAI has launched Structured Outputs, but what does it mean to have structured data output? ✅ more predictable outputs ✅ no more parsing, less output validation ✅ simpler integration ✅ one API call, but with all the info we need ✅ more reliable output, i.e. you can build an assistant in a more reliable way All OpenAI models are available? only few model available, but OpenAI included newer GPT-4o snapshot and after few tests is very promising GPT-4o newer snapshot achieves 100% reliability in matching output schemas How?  You can use it in 2 ways: Function calling Response format: you can pass the structure you want Just add response_format parameter with the schema you want and the option “strict”: true. 🚀 #OpenAI #LangChain #StructuredOutputs #AI #OpenAIAPI #JSONSchema #APITutorial #MachineLearning

Microsoft is venturing further into AI with the development of a new language model, MAI-1, as it seeks to compete with giants like Google, Anthropic, and OpenAI. The development of MAI-1 follows Microsoft's significant $10 billion investment in OpenAI and the acquisition of Inflection's team and IP for $650 million. Led by former Google AI head Mustafa Suleyman, MAI-1 is set to be a formidable contender in the AI arena with around 500 billion parameters, aligning it closely with OpenAI's GPT-4. Microsoft plans to introduce MAI-1, which includes unique techniques from the Inflection team, at their upcoming Build conference, hinting at new, diverse applications for both mobile and cloud-based environments. #MicrosoftAI #TechInnovation #AIdevelopment #AI

OpenAI being closed-source forces it to outpace open-source competitors and give consumers a reason to pay. In my view, this is driving rapid, excess innovation across the industry (e.g., o1 model). By tackling fundamental infrastructure challenges and model constraints like reasoning & reliability, OpenAI is setting the bar high, creating a trickle-down effect across the ecosystem. Enterprises and startups alike can leverage this foundational progress to focus on building smaller, targeted applications - like small language models (SLMs) - that are more efficient and domain specific. #AI #OpenAI #Innovation #LLMs #SLMs #AITrends Source: Artificial Analysis

 "Pablo Villalobos, who studies artificial intelligence for the research institute Epoch, estimated that GPT-4 was trained on as many as 12 trillion tokens. Based on a computer-science principle called the Chinchilla scaling laws, an AI system like GPT-5 would need 60 trillion to 100 trillion tokens of data if researchers continued to follow the current growth trajectory, Villalobos and other researchers have estimated. Harnessing all the high-quality language and image data available could still leave a shortfall of 10 trillion to 20 trillion tokens or more, Villalobos said. And it isn’t clear how to bridge that gap.  Two years ago, Villalobos and his colleagues wrote that there was a 50% chance that the demand for high-quality data would outstrip supply by mid-2024 and a 90% chance that it would happen by 2026. They have since become a bit more optimistic, and plan to update their estimate to 2028. " #changeonsdeperspectives KIF-KnowledgeIMMERSIVEForum https://lnkd.in/e8RGBrim

As I keep testing OpenAI's highly-anticipated model, O1 (Strawberry), which has been touted for months as their most advanced creation to date, I was left with more questions than answers. Based on model's behaviours, my initial impression is that O1 might not be a standalone model. Instead, it feels more like a chain of thought built on top of GPT-4o. If my assumption is correct that O1 is not a novel model, but rather a chain of thought using GPT-4o, then it raises questions about the current state of the Large Language Model (LLM) industry and suggests that the LLM development might have hit the wall. This is a good thing though as it reinforces my long-standing hypothesis that multi-agents and chain of thoughts will be the future of AI. I believe that the key to unlocking the next level of AI advancements is the use of multiagents and chain of thoughts, which allows for more flexible, adaptable, and human-like reasoning and a crucial step towards achieving Artificial General Intelligence (AGI). The future of AI is not just about scaling models, but about creating more sophisticated, dynamic, and collaborative systems. #ArtificialIntelligence #LLMs #OpenAI #AGI #AIIndustry #Innovation #O1

The amount of new releases and progress of the open source AI community is just blowing my mind. Never ever have I witnessed such a level of open innovation, with so many participants. OpenAI, Anthropic and Google are becoming lonely defenders of the dark world of proprietary LLM's. Today, Microsoft released Phi-3 - #phi3-mini: 3.8B model trained on 3.3T tokens rivals Mixtral 8x7B & GPT-3.5 - #phi3-medium: 14B model trained on 4.8T tokens w/ 78% on MMLU and 8.9 on MT-bench

🚀 Welcoming update from OpenAI! They just introduces GPT-4o mini—a smarter, cheaper, and just as fast model as GPT-3.5 Turbo! 🧠💡 -- Outperforms GPT-3.5 Turbo in intelligence (82% vs. 69.8% on MMLU) -- Over 60% cheaper: $0.15 per 1M input tokens, $0.60 per 1M output tokens -- Supports text and vision, with audio and video coming soon -- Improved multilingual understanding Perfect for high-volume, cost-sensitive, and fast-response tasks. 📈🔍 #OpenAI #GPT4oMini #AI #MachineLearning

Live updates below: GPT-4o is the new model announced! - FREE (paid users will get higher rate limits and other features) - Truly/natively multimodal, with what seems to be near real-time speech-to-text, no longer need to wait to interrupt it, emotionality sensing and generating - Advanced Data Analysis - Available via API - Will be rolled out to users shortly Around 19h00 SAST this evening we will know what OpenAI plans on unveiling. Will it be GPT4.5 like some anticipate, or lower token costs and new tools for GPT4? #openai #gpt #ai #llm #gpt4o

Companies are increasingly deploying smaller and midsize generative artificial intelligence models, favoring the scaled down, cost efficient technology over the large, flashy models that made waves in the AI boom’s early days. Unlike foundation models such as OpenAI’s GPT-4—which cost more than $100 million to develop and uses more than one trillion parameters, a measure of its size—smaller models are trained on less data and often designed for specific tasks.