ChatGPT is experimenting with an option to continue shared conversations. Again.

New NotebookLM Audio Overviews are quite impressive.

It turns any content into a podcast in a few seconds 👀

Google rolled out Audio Overview feature for NotebookLM
https://www.testingcatalog.com/google-rolled-out-audio-overview-feature-for-notebooklm/

🗞 #ai

Google rolled out Gems and Gem Manager support to mobile apps
https://www.testingcatalog.com/google-rolled-out-gems-and-gem-manager-support-to-mobile-apps/

🗞 #gemini

ChatGPT is rolling out system shortcuts support with /picture and /search options.

Users can now type / to see a pop-up selector with available shortcut options. These selections will trigger a relevant chat experience based on your selection.

The Picture option will trigger image generation with DALL-E even in case you do not specify any particular prompts.

The search option will open a new Search mode where every prompt you enter will be used as a search query. Interestingly, if you don't specify any prompt it seems to be taking the last note from the memory.

Not an update you are all waiting for but ChatGPT got new 404 page descriptions which are quite extensive 👀👀👀

I now see conspiracy and hidden messages everywhere 😅

🚨 BREAKING: OpenAI finally announced a new o1 model with better reasoning!

“These enhanced reasoning capabilities may be particularly useful if you’re tackling complex problems in science, coding, math, and similar fields”

Source

And evals 👀👀👀

Looks insane tbh. But needs some testing first!

Got my o1 preview access as well. Be aware of strict usage limits and test wisely 🤖

Other models are now available in the "More models" section. Also now there is a new system shortcut for "Reasoning"

ChatGPT o1 prompt guide to keep in mind for your 50 weekly messages limit 👀👀👀

LoL, how many rs in strawberry is now a suggested conversation starter 😅

ICYMI: ChatGPT shows blue dots on the conversation history in case you left the conversation before the response was finalised.

These are notifications of completion 👀

Do you also hoard your compute quota for o1 model and plan to utilize it on something really useful?

I will invest it in news publishing workflows to evaluate its writing capabilities 👀

I haven't run a lot of testing yet but here are some quick observations:
- Retrieving knowledge via search didn't change much - o1 blindly trusts the Bing search index unless you clarify your reasoning.
- Written outputs follow the same GPT writing style and are not "humanised" at all.
- However, if you ask it to "humanise" the text - it does an impressive job.

As a next step, I am planning to test more complex writing prompts 👀👀👀