Review of 12 Days of OpenAI
Posted on:
I wanted to comment on each of the 12 days of products and announcements that OpenAI put out this December.
Day 1: o1 & ChatGPT Pro
$200 a month for individuals is a large reach, but I could see it be beneficial as what you get with this Pro plan increases. The "reasoning" model being out isn't really too exciting for me given it is just a model and not really a product.
Day 2: Reinforcement Fine-Tuning
Fine tuning a model towards specific categories or topics I think will be useful down the road, but again this is a research feature and not a product, so what comes out of this is still in the future.
Day 3: Sora
Text-To-Video is definitely a powerful tool, but it is scary when people who aren't fluent with media literacy see a generated video and think it is real. I get why OpenAI wants to make this, but in an era of misinformation, this is putting more gasoline on the fire without telling users how to be safe with this new tool.
Day 4: Canvas
This on the other hand I do think is going to be more beneficial. A back and forth of working on documents or code where you have control and can tweak things without having to prompt ChatGPT every time or ask it to just change a highlighted section is much lower risk and higher reward for end users.
Day 5: ChatGPT in Apple Intelligence
Apple Intelligence currently is a lot of gimmicks and party tricks overall so far as of December 2024, but I have commented on this previously that making Siri more capable even if it is just reaching out to an external service is a step in the right direction.
Day 6: Advanced Voice & Santa Mode
Definitely a small low risk and fun thing that they added with the santa voice. That said, being able to use Video in their voice mode is an actual feature that is pushing the tool forward.
Day 7: Projects
Folders. Thank god.
Day 8: Search
As I said in my Weeknotes this week, having ChatGPT cite sources is I feel a great step forward, but if only it was across the system and not just here (I know that is unlikely given how the tech works, but maybe take the output of a query and search it to validate what ChatGPT outputs is accurate).
Day 9: o1 via OpenAI API, Realtime API improvements
Useful for people who are integrating OpenAI models into their apps and software, but I'm not one of those people so I'm neutral on this.
Day 10: 1-800-CHATGPT
Kinda a party trick and I gotta wonder how long this actually will last.
Day 11: Work with apps
Actually getting access to your tools rather than just screenshots I think can be useful, as long as the data being shared with OpenAI is secure, not being used to train their models further, and doesn't snoop around past where you tell it to. That said, this is a step further towards agents, but at the same time, it is still not quite there as it currently isn't actually using your apps, so it is a read-only path right now rather than full read/write.
Day 12: o3
As I said previously, it's a new model, but almost nobody has access to it so until it is actually used by users and built into apps, it is just a research space for now.
Conclusion
I do think they did push out a bunch of good features over the past few weeks that pushes ChatGPT's capabilities forward, but now I feel it is up to both users and devs to actually use those capabilities. As with the AI space at large, it is still fast moving and you have to have some skepticism in the actual usefulness, how accurate the tools are, and if this is bringing value to users and not just the venture capitalists throwing billions of dollars into this.