AI can now REASON?! tl;dr: no, it cant!
Another day, another AI model drop! This time, it’s the OpenAI o1 series, and wow, the hype is all over my feed 🙄 Is this the breakthrough in reasoning we’ve all been waiting for? 🤔
OpenAI claims these models are designed for coding, math, and science. Supposedly, they’re better at reasoning through tough problems. But wait, what?! They aren’t actually reasoning. What they are doing is mimicking reasoning using a technique called “chain-of-thought” processing. How this works? Instead of jumping straight to an answer, these models break down a problem step-by-step, processing more data (tokens) and taking longer to respond. This mimics how humans reason through problems, but at the end of the day, it’s still advanced pattern recognition – the model is just generating text based on its training, not “thinking” like we are 💡
In theory, this should lead to more accurate outputs in areas like coding and math, where getting things wrong is a big issue. But there are trade-offs:
- Reduced context windows due to the model using tokens for “reasoning steps”
- Longer wait times for responses (something between few seconds to few minutes)
- And yes, more money – these models are far more expensive to run
For now, they’re only in preview (with very strict usage limits) and limited to text-based tasks.
To me, the bigger takeaway is the shift toward specialized models. Instead of trying to make one model do everything, OpenAI is focusing on building models for specific, high-demand tasks like coding, math, and science. Maybe this specialization is the real innovation here.
What do you think?
Published on:
Learn moreRelated posts
You are holding GitHub Copilot Wrong!
Most developers think that one can’t really use GitHub Copilot wrong. There is a chat interface that lets you also choose a model, so th...
You are holding GitHub Copilot Wrong!
Part 0 showed why constant prompting, re-prompting, and steering GitHub Copilot feels fragile. Not because Copilot is unreliable, but because ...
Building a Multi-Hierarchy Ticket Classification System (Because Keywords Aren't Enough)
Look, I love a good keyword-based system as much as the next developer. They’re fast, predictable, and when your user says “VPN,&r...
Secretless cross-tenant dataverse access
Client secrets are like hiding your house key under the mat; easy to grab and impossible to audit. Certificates are just slightly better, beca...
How Azure CLI handles your tokens and what you might be ignoring
Running az login feels like magic. A browser pops up, you pick an account, and from then on, everything just works. No more passwords, no more...
How Dev Proxy teaches you to make your apps more resilient
I added Microsoft Dev Proxy to my Mermaid → Dataverse converter, because I wanted to test how it handled rate limits and API errors. What I go...
Building Azure functions that never store secrets — ever
What if your function could hit Microsoft Graph with no client secrets, no certs, and no Key Vault entries? That is exactly what a Managed id...
Introducing Mermaid to Dataverse Converter
Why diagrams matter (and why they usually fail us) Entity-Relationship Diagrams (ERDs) are the universal shorthand for talking about data mode...
It’s OK to be seen trying
It’s OK to be seen trying Somewhere along the way, we started believing that you’re only allowed to speak after you’ve figured everything out....
Stuck in pilot - Part 1: no foundations, no future
“We just need to test the AI. We’ll figure out the data later.” That sentence has quietly killed more AI pilots than any model failure ever ...