Kaushik Shakkari’s Post

📢 Breaking News: OpenAI Unveils o1 Preview! 🧠💡 OpenAI announced the release of their groundbreaking o1 series - AI models designed to think longer and deeper before responding. Key highlights: 🚀 Available now for ChatGPT Plus and Team users, and API developers (tier 5) 🧪 Excels in complex reasoning tasks for science, coding, and math 🏆 Outperforms PhD students in physics, chemistry, and biology benchmarks 💻 Reaches 89th percentile in Codeforces coding competitions 🥇 Scores 83% on International Mathematics Olympiad qualifying exam (vs. 13% for GPT-4) 🛡️ Enhanced safety features: New training approach leveraging reasoning capabilities Improved resistance to "jailbreaking" attempts (o1-preview scored 84/100 vs. GPT-4's 22/100) Rigorous testing using Preparedness Framework Best-in-class red teaming and board-level review processes 🌐 Government collaboration: Formalized agreements with U.S. and U.K. AI Safety Institutes Early access granted to research version for evaluation and testing 👥 Potential applications: Healthcare research (e.g., annotating cell sequencing data) Quantum physics (generating complex mathematical formulas) Advanced software development (multi-step workflows) ⚙️ Technical insights: Uses large-scale reinforcement learning to enhance thinking processes Performance improves with more training compute and thinking time Exploring new scaling constraints different from LLM pretraining 🚧 Current limitations: Lacks some features of GPT-4 (web browsing, file/image uploads) Still in preview, with regular updates expected 💰 Pricing: $3.00 per 1M input tokens This release marks a significant leap in AI capabilities, potentially revolutionizing complex problem-solving across various fields. OpenAI is resetting the counter to 1 with this new series, signifying a new era in AI development. What potential impacts do you foresee from this advancement in AI reasoning capabilities? How might it affect your industry? Share your thoughts below! 👇 #OpenAI #AI

"Strawberry" is here! 🍓 The new model from OpenAI can reason independently to solve more complex tasks. It's a new scaling paradigm that, instead of creating larger models, focuses on inference compute to question itself "behind the scenes" to produce better results.

For PlatorAI followers - has anyone with Plus or API use used this yet? There was a slight lag of between announcement and availability - it’s now available on our App* (except advanced voice - see other post!) I would be interested to find out users initial views. We will be exploring its use cases on both Plus App and API over the coming weeks. #PlatorAI #EthicalAi #OpenAi

OpenAI o1 marks a crucial advancement in AI, enhancing its ability to reason through complex tasks. This development highlights the importance of ethics in using AI to tackle future challenges in science and technology.

Exciting times in AI! We’re seeing a monumental shift as models evolve from fast, intuitive responses to deep, deliberate reasoning—transitioning from System-1 to System-2 thinking. This leap enables AI to tackle complex problems in science, coding, and mathematics that were once out of reach. By embracing chain-of-thought reasoning, AI now spends more time thinking through problems before responding, refining strategies, and recognizing mistakes—much like we do. This not only enhances accuracy but also reduces errors and misunderstandings. The rapid pace of AI development is shortening the shelf-life of internal tools, making adaptability more crucial than ever. Can’t wait to see where this evolution takes us!

AI wrote the following comment for me: “Great!”

Finding the limits of GPTs over the last 18 months has been an evolving passion in #promptEngineering but not distinctly different from a form of algorithmic or #programaticDevelopment in itself. Despite the time saving benefits, 5-6 levels of abstraction depth encounters many #LLMModel #InterpretiveChallenges even in coding from templated applications. More capable complex models with Agents will be a defining moment in where AI will take us in Applied use in enterprise and solving bigger problems. #excitingGPTDevelopments

This is really going viral. OpenAI released its new model #o1 with reasoning. Now that the storm 🌬 has calmed down a little overnight from a European perspective and people have started to think for themselves, here's what I think: - OpenAI o1 is not ‘the’ #AGI that many expected. It is much more a further development to correct known weaknesses. - The main performance improvements are, as OpenAI confirms, due to the reasoning part. A good drawing from Tom Yeh of how this is done after the actual #LLM can be found here: https://lnkd.in/eFhSxqJR - I think OpenAI o1 is proof that LLMs do not get better through scaling, but rather through an approach that is comparable to that of automatic agents. However, it remains to be seen whether the reasoning core, insofar as it is integrated into the model, will perform better in the long term. My prediction: #OpenSource LLMs will close the existing gap with something like automatic agents #AutoAgents in a short time. My conclusion: It will be increasingly important to execute LLMs quickly in order to be able to iteratively improve a result in a reasonable amount of time. In the future, we will consume more tokens for a request in order to achieve better results. What do you think? 🤔

💡 Reflection of the Day 💡 Very impressive, AI models like GPT-o can process massive amounts of information, but if no human can manage that volume, who ensures its answers are accurate, especially in fields like genetics? This raises an important point about the role of humans in guiding and assessing AI. AI models are powerful tools that can help us work faster, but to ensure both speed and quality, we need professionals who can lead, evaluate, and improve these tools continuously. Growth as professionals is essential as AI becomes more integrated into our work. #AI #Genetics #ProfessionalGrowth #AIEthics #TechnologyLeadership

I cannot wait to see where the world of medicine & healthcare goes with the implementation of AI.