Ray Villalobos’ Post

New AI Model - O1 from OpenAI. Shows improvements in reasoning by letting the model 'think' for longer. Seems cool for sure, but this isn't GPT-5 or a generational breakthrough for most users. If you have a specific, difficult problem that AI hasn't been able to solve for you yet, it's worth trying out on O1. Lots of different usecases that were previously just out of reach are probably in striking distance. In fact, it seems to me that one of the biggest uses of O1 will be in performance on high-context understanding of complex business problems. But you're so far from being ready to actually leverage that! We still need lots more exercise and exposure with thinking about these tools before we're going to be able to - as a practical and behaviour and people matter - roll these tools out at scale. So it's a neat development! Try it out! Experiment! Play! Have fun! Doesn't really matter what model you're using as long as you're pushing to get your experience in.

OpenAI's latest advancements with the o1-preview and o1-mini models are groundbreaking! 🚀 With o1 achieving PhD-level accuracy in STEM fields and ranking among the top in coding competitions like Codeforces, we're witnessing AI excel at tasks that were once exclusive to human expertise. This opens doors to massive possibilities, from revolutionizing how we approach complex problem-solving in science and engineering to enabling faster, cost-effective AI applications with o1-mini

💡 OpenAI’s New o1 Models: Powerful, but Expensive – Use with Care! OpenAI has just introduced two new models, o1-preview and o1-mini. While these models seem impressive, they come with a hefty price tag that may not fit every use case. 🔍 Pricing Breakdown: o1-preview $15.00 / 1M input tokens $60.00 / 1M output tokens o1-mini (more cost-efficient but still pricey!) $3.00 / 1M input tokens $12.00 / 1M output tokens ⚠️ My POV: These models are likely expensive for a reason and could be worth the cost in high-value scenarios. But that doesn’t mean it’s time to start replacing every model in your stack with o1-preview just yet. If your organization relies on LLMs to make decisions in high ROI situations, then these models might justify their price. However, for everyday tasks like intent recognition or LLM routing, swapping out your current models for o1-preview could lead to massive overhead without the proportional benefit. It’s all about choosing the right tool for the job! 💰🔍 The new models have great potential, but before making the leap, evaluate the cost-effectiveness carefully. Let’s not fall into the trap of using the latest tech where it’s not needed. Curious to hear if anyone else is planning to adopt these models—what’s your strategy?

Wow. It’s getting deeper and deeper. We saw that in. 1980 as the computer power increases. Now it happens again. But more in the direction of AI. The next generation of AI models are designed to [think] more before they answer.

This is getting interesting.. Quality and hallucinations are one of the biggest concerns of the current #GenAI tools. #edtech #elearning

The biggest technical challenge for GenAI companies is building models to meet their lofty promises and big dreams. GPT-o1, nee "Strawberry" is an example of where I think we're headed next: More complex models built to simulate advanced "reasoning" through multi-step processes and more compute. Read between the lines and this starts to sound a lot like OpenAI serving a multi-step fine-tuned CoT system, heavily filtered and indexed through RLHF, as a model. That helps explain - limited access (enterprise, edu, and tier 5 only for the API, 30 prompts per week for ChatGPT Plus and Teams users), - 128k context window, some of which is consumed by new "reasoning tokens", - significantly slower responses ("anywhere from a few seconds to several minutes" according to documentation) - limited abilities (text only - no multimodal, no tools and function calling, no streaming) - specialized use cases - OpenAI recommends using 4o1 and 4o1-mini for tasks that require "more advanced reasoning" - specifically coding, math, and science problems, - anticipated higher cost due to compute demand This model just dropped, and it's not in my hands yet so I can't say anything meaningful about the actual performance. Once I have access I'll do comprehensive testing and give you my take on what this model is for and how to build solutions with it. More to come! #openai #strawberry #gpt #gpt4 #gpt4o #gpt4o1 #4o1

OpenAI o1-preview and o1-mini are rolling out today in the API for developers on tier 5. o1-preview has strong reasoning capabilities and broad world knowledge. o1-mini is faster, 80% cheaper, and competitive with o1-preview at coding tasks. Open AI posted this announcement 58 minutes ago and same time it was announced it is available on Azure: Introducing o1 (link doesn't open in Safari but will only open in Ms Edge!) : https://lnkd.in/e5yyC8HM Who am I say to judge what pace this is moving? But then again we needed the model which spends more time thinking before responding, while ago! Better late than never...

Our engineering team just got access to OpenAI's new model, launched today. We’re thrilled to dive into this latest innovation and explore how it can push the boundaries of what we’re building. Big thanks to OpenAI for continually driving the AI frontier forward. Here’s to making technology work for good! 💪 Check out OpenAI’s original announcement below to learn more about their new model. 👇 #AI #Innovation #TechForGood #UnshackleAI #OpenAI

Revolutionizing AI: The Power of 'Thinking'! OpenAI 's latest innovation, o1, is changing the game with its unparalleled ability to 'think' before responding. This cutting-edge technology enables: Multi-step reasoning Contextual understanding Large language model capabilities Meta AI architecture With o1, AI can now: Evaluate information like a human Identify patterns and make logical connections Provide more accurate and detailed responses Imagine the possibilities: Enhanced problem-solving Improved decision-making Streamlined content creation and many more. Stay ahead of the curve and explore the potential of o1 today! #AI #OpenAI #Innovation #Technology #FutureOfWork"

Took OpenAI O1 a spin with latest leetcode weekly contest’s hard problem. https://lnkd.in/gbX74W-x , definitely amazed by it’s ability to reason and execute the game plan. It was able to identify the problem setup, mimicking a chess game, identifying it is best solved with DP, and generate Python code (even with some bit manipulation hack to optimize for compute, although it’s a bit unnecessary and sacrificed code readability). Not to discuss how advanced the AI system is and technical detail of O1, since technical Claude has been excellent with coding tasks. But one thing for sure companies need to realize is using leetcode to test candidates’ quality is getting more pointless than ever. It is a solution to hire at scale, but companies really need to find other form of interview processes to hire for quality.