The Information explores how OpenAI's reasoning models could help counteract potentially slowing progress from pretraining of larger more sophisticated models with more incremental gains over earlier models. It also notes that "relatively few developers" are using o1 models. While I haven't seen the full-fledged o1 models (and am certainly not a developer), the current models sometimes think hard about how to break code -- have seen them replace functional authentication with placeholders that you have to go back and update later, or swap out the correct variables with wrong ones -- for instance upper- or lowercasing something you didn't ask for, or adding an underscore where you didn't want one. At present, I think it still is better to have tighter feedback loops with the user -- otherwise, the genAI is thinking long and hard -- while a game of telephone can ensue where mistakes sometimes increase in the process of reflection -- about something that might not be what you wanted. A snippet from the paywalled article. The Information has done a lot of unique reporting on the subject.
Brian K. Buntz’s Post
More Relevant Posts
-
OpenAI's o1 model and its pro mode enhance performance relative to their predecessor, o1-preview. In particular, o1's pro mode boosts computing power at inference to improve the model's ability to solve hard problems. Currently only accessible via a $200 monthly ChatGPT Pro subscription, o1 supports text and image inputs and delivers more accurate outputs than o1-preview or GPT-4o, particularly when performing complex reasoning tasks. The o1 pro mode delivers better results than o1, especially in repeated trials, but requires more processing power, leading to higher costs and slower responses. Learn more in #TheBatch: https://hubs.la/Q02_BZrM0
-
OpenAI has introduced a powerful new feature in their API: Structured Outputs. This enhancement allows developers to enforce strict JSON schemas, ensuring that AI-generated content is both accurate and precisely aligned with required formats. With Structured Outputs, OpenAI models can now seamlessly generate complex data structures, reducing the need for repeated requests and manual adjustments. Techniques like Constraint Decoding ensure every token adheres to the specified schema, providing unmatched reliability. This feature marks a significant advancement for developers looking to create robust, error-resistant applications. Explore the full details and see how Structured Outputs can elevate your AI projects: https://lnkd.in/d5PA7HGQ
How OpenAI implements Structured Outputs
newtuple.com
-
Pond & Assoc. - People, Process and AI - References upon request. Schedule a free consultation - see link below.
Read the prompt below. The question posed is not simple addition or complete a sentence. The prompt is asking ChatGPT to 'reason', to solve a puzzle. It is September 2024. Open AI released a tool that can operate at a high human levels. Buckle up. Play with these tools. Learn now. Be ready for what's coming. #ai #genai Paul W. Kevin Rank, MBA
Follow me to learn how you can leverage AI to boost your productivity and accelerate your career. Scaled products to 8 Million+ users and built the world's biggest AI newsletter with 1 Million+ readers.
Nothing to see here. Just OpenAI's new model showing human level reasoning abilities. This would take most people at least a couple of minutes to solve. The new o1 model did it in 15 seconds.
-
-
Everyone will talk about OpenAI o1 model. This video is a great summary of real examples what is improved. Recommend to watch it. Mainly if you already use structure prompting, nothing will change much, but in general inputs the chain-of-thoughts are shown. https://lnkd.in/d4de8TiU
OpenAI Releases Smartest AI Ever & How To Use It
https://www.youtube.com/
-
In case you missed it last week... Microsoft released Phi-3, a small 3.8B language model 💡 If you're expecting out-of-the-box OpenAI-like performance from it, you'll be disappointed. However, this is the best small language model (AKA SLM) I've seen so far, with a 128k context window, and you can run it on any device with good latency. If you have a very concrete use case, this is a solid candidate for a cost-efficient model to use after fine-tuning. 📚 Links: Phi-3 on Hugging Face - https://lnkd.in/dYgAyzpd Original blog - https://lnkd.in/dHpMvWHG Paper - https://lnkd.in/dDmVvbWC Playground - https://lnkd.in/dT7_-Y4U
-
-
It's important that we apply a more considered and objective thought to posting on the merits of GenAI, especially when benchmarking with human reasoning. As per earlier discussion with TI (PhD and Multi-Mastered), LLMs operate probabilistically, where what might appear as "reasoning" is actually the model generating diverse responses based on probabilistic outcomes. This variability is why repeated prompts can lead to different answers. True reasoning, as humans understand it, involves consistent logic and comprehension, which LLMs do not replicate in the same way. https://lnkd.in/gShSjkiT
Follow me to learn how you can leverage AI to boost your productivity and accelerate your career. Scaled products to 8 Million+ users and built the world's biggest AI newsletter with 1 Million+ readers.
Nothing to see here. Just OpenAI's new model showing human level reasoning abilities. This would take most people at least a couple of minutes to solve. The new o1 model did it in 15 seconds.
-
-
Exciting news from OpenAI! On January 23, 2025, they introduced "Operator", an AI agent designed to handle tasks like ordering groceries and booking travel. Initially available to ChatGPT Pro users in the U.S. for $200/month, yet OpenAI plans to expand access and eventually integrate "Operator's" capabilities into ChatGPT for more users. As someone passionate about data analytics, I can’t help but wonder how game-changing "Operator" could be in our field. The potential for automating routine tasks, streamlining workflows, and uncovering insights faster is incredible! Instead of trying to put my excitement into words, I’ll just let this video do the talking. Take a look - I promise it’s worth your time! 🎥🚀
I Let OpenAI Operator Manage My Google Sheet: Here’s What Happened.
https://www.youtube.com/
-
Microsoft suggests that OpenAI’s o1 Mini and GPT 4o Mini consist of 100 billion and 8 billion parameters, respectively. https://hubs.li/Q031mFPJ0
Did Microsoft Spill the Secrets of OpenAI?
-
Remember that $2000 OpenAI model I talked about? Well, they released it! No, they don’t charge 2k for it yet. But I think it can actually be worth it. The model is called o1. The huge difference is that it actually thinks before answering. It means o1 can solve problems that it hasn't been trained on yet. With this chain-of-thought approach, the model crashes its predecessor GPT4o by a big margin. Charts just speak for themselves 📈 ↳ Scroll the carousel and be amazed. We are slowly getting closer to AGI. So excited about the next couple years. P.S. Want to learn about the JTBD framework tomorrow? Join our newsletter! It’s free and has tons of value. Click the link in the comments to join.
Editor-in-Chief, R&D World @ WTWH Media LLC | Data-Driven Storyteller
3mohttps://www.theinformation.com/articles/goodbye-gpt-hello-reasoning-o?rc=rteufi