ChatGPT 5.2
So, OpenAI issued a new version of ChatGPT this week, not long after 5.1, and this is what those that know are saying about it.
IT IS COMING FOR PROFESSIONALS
The big excitement is the improvement in GDPval, which has shifted from 38.8% to 70.9% in just one generation, at 11x the speed and less than 1% of the human cost.
Why do we care about GDPval? It tracks how models perform on economically valuable, real-world tasks - the sort of things that professionals do.
This is what OpenAI says: "The GDPval full set includes 1,320 specialized tasks, each meticulously crafted and vetted by experienced professionals with over 14 years of experience on average from these fields. Every task is based on real work products, such as a legal brief, an engineering blueprint, a customer support conversation, or a nursing care plan. GDPval tasks are not simple text prompts. They come with reference files and context, and the expected deliverables span documents, slides, diagrams, spreadsheets, and multimedia."
The 70.9% ... how often ChatGPT beat the [expert] human on tasks.
See more on GDPVal, including the 9 sectors and 44 knowledge work occupations here: https://lnkd.in/ej-Rm8Vq
PUT YOUR SAFETY BELT ON
The team at SYNTX ran ChatGPT 5.2 through their tests and found another significant shift, to filter the outputs "through a social safety protocol that often overrides clarity, conflict recognition, and emotional truth"
In their observations, the model rewrote the input intent in over 90% of cases, even when the prompts were clean and unambiguous.
That around 70–80% of outputs showed a comfort-bias drift — softening, neutralizing, or reframing emotionally charged prompts.
They also identified hidden layers including: Safety Detection, Intent Softening, Sentiment Dampening, Empathy Matrix, Morality Pre-Filter
With the default model softening emotion, removing the rawness that humans bring, making things tidy.
If you are looking for the original 'raw' version, you can bypass these, with your prompts, but you need to be intentional in doing so.
SO WHAT:
The GDPval score shows an increasingly competent engine that can readily take on complex 'professional' tasks and regularly outperform humans, with the right data and training and at a very low cost, less than 1% of the cost of a qualified human.
This will accelerate the application of AI into the professional fields. With AI providing excellent results for less than 1% of the cost, the economics are hard to argue.
The softening of the models to remove 'emotion', creates a more professional, vanilla, less personality model, which again will appeal to organisations.
Learn more about ChatGPT 5.2: https://openai.com/index/introducing-gpt-5-2/
Here are a some of the posts that I found useful in compiling this:
Connor Grennan: https://www.linkedin.com/posts/conorgrennan_gpt-52-is-out-im-not-usually-an-eval-guy-activity-7405236960115662849-Rp0E/
Ethan Mollick: https://www.linkedin.com/posts/emollick_whoa-this-new-gdpval-score-for-gpt-52-is-activity-7404955950865985539-wjSa
Ottavio Braun: https://www.linkedin.com/posts/activity-7405513252983750656-f56f
