Refresh
Of course, OpenAI had to bring us back down to earth with another joke, though this time, we didn’t get to hear Sam Altman laugh at it, though. With Christmas fast approaching and the business being based in San Fransisco – home to many a self-driving car – it’s pretty on point.
The joke went along the lines of: We live in San Fransisco, self-driving vehicles are all the rage, and Santa’s been trying to get in on this. He’s been trying to make a self driving sleigh, but it keeps hitting trees left and right. Any guesses? He didn’t pine-tune his models.
To help you better visualize it, TechRadar’s Editor-at-Large Lance Ulanoff asked ChatGPT to create an image of it.
While this is not necessarily consumer-focused, OpenAI welcomed Justin Reese, a researcher for rare genetic diseases at Berkeley University. Now, in a live demo on o1 mini, we’re trying to have the model ID genes from a sample data pool that might be responsible for a disease. Of course, it will be fact-checked against known results, but it’s a good test of reinforcement fine-tuning for validation on o1 mini.
Furthermore, within o1 mini, you can refine and customize this fine-tuning process. The aim is to let you get the most out of the data by tweaking the model to your specific needs. It was noted that depending on the amount of data and the task, it can take anywhere from a few seconds to hours to complete.
The results at the end will be plotted within an evaluation dashboard with several presets.
Day 2 of ’12 Days of OpenAI’ is now kicking off, and Sam Altman is not here. Instead, some other team members are teasing a tool that will arrive next year. It’s not a new model or anything in the realm of Sora, but OpenAI’s customization for models will now support reinforcement fine-tuning.
One of the nice things about Day 1 of OpenAI’s “12 days of OpenAI” was its brevity. When Sam Altman sat down before us with three of his engineers, we wondered it we were in for hours of exposition on major AI updates across every aspect of OpenAI’s business. Turned out that OpenAI was really spreading out all its news over 12 days. Yes, that means we’re in for a lot over the next week or so but at least we can count on each day of news being digestible.
On the other hand, can we get to that Sora update and release today? Please? We’re just 10 minutes from finding out…
Here’s another prediction for “12 Days of OpenAI” announcements: videochatGPT. You might be familiar with the Norad Santa Tracker It’s a fun way to track Santa’s flight around the world on the 24th of December. This year I’m thinking that it’s the perfect time for OpenAI to take this a step further with an AI Santa video call demo as part of a new AI video ChatGPT.
So, imagine talking to ChatGPT just like you’re currently doing in ChatGPT Advanced Voice Mode, but you’re seeing a video avatar talk back to you. Santa Claus would be the perfect video avatar to kick this off, and it would delight children everywhere. Let’s just hope AI Santa doesn’t start hallucinating because the result could be hilarious…
I mentioned Sora earlier, but just in case you haven’t heard of the AI video generator, here’s an unbelievable trailer from OpenAI showing just what it’s capable of.
My personal favorite here is the prompt “a litter of golden retriever puppies playing in the snow. Their heads pop out of the snow.”
How can AI be so cute? I can’t quite wrap my head around it.
ChatGPT uses Dall-E 3 for image generation, which is right up there with the current crop of AI image generators on the market, however, it’s starting to look a little long in the tooth. New upcoming AI image generators like Flux have been slowly getting better and better.
Could we see a new version of Dall-E in our 12 days of OpenAI? It’s a hotly tipped possibility. If OpenAI can give us image generation that’s better than Flux Pro then it will certainly be a happy holiday season for everybody.
OpenAI o1 is now out of preview in ChatGPT.What’s changed since the preview? A faster, more powerful reasoning model that’s better at coding, math & writing.o1 now also supports image uploads, allowing it to apply reasoning to visuals for more detailed & useful responses. pic.twitter.com/hrLiID3MhJDecember 5, 2024
If you’re just joining us, OpenAI announced the official launch of o1 yesterday, with a “faster, more powerful reasoning model that’s better at coding, math & writing.”
If you want to try it out yourself, just head to ChatGPT and choose the o1 model from the dropdown. Give it a try on your math homework, or a coding challenge, you might be surprised by the results.
OpenAI’s next livestream kicks off at 10 am PT / 1 pm ET / 6 pm GMT and you can watch it live directly from OpenAI’s website.
If you go to the “12 days of OpenAI” section of the website right now, you’ll see a gorgeous advent calendar, hinting at the exciting days to come. Just like an advent calendar, some days will be better than others, so I’m very intrigued to see if OpenAI keeps up the momentum or if today isn’t quite as stellar as yesterday.
Bookmark that link too, as you’ll be able to revisit all the highlights from the events even if you miss one of the livestreams. Or, you could keep checking in with TechRadar as we’ll keep you up to date on everything you need to know over the next week or so.
What are TechRadar’s predictions for day 2 of OpenAI’s “12 days of OpenAI” I hear you cry? Well, Sam Altman said that we can expect “some big ones and some stocking stuffers” throughout the 12 days, and considering that yesterday we got a brand new version of ChatGPT (ChatGPT o1) I’d predict that today’s Xmas gift from OpenAI will be more of a “stocking stuffer” than one of the “big ones”.
Perhaps a minor update to ChatGPT search or ChatGPT Advanced Voice Mode. What am I really hoping for? I want ChatGPT search to be rolled out to all users on the free tier. Fingers crossed!
One of the biggest announcements we expect to see over the next week or so is the official launch of Sora, OpenAI’s video generator which can transform a text prompt into an incredible video.
Sora was leaked last month by unhappy artists who have accused OpenAI of taking advantage for unpaid research and development purposes. There’s definitely a debate to be had on how OpenAI trains its AI models, but that’s maybe one for another day.
In terms of what Sora offers, well, imagine one of the best AI image generators, but video. I’ve not tried Sora yet, but from the demos online, it looks pretty awesome.
OpenAI also announced ChatGPT Pro yesterday, but who is it actually aimed at?
For $200/month ChatGPT Pro gives you unlimited usage and an even smarter version of o1 with “more benefits to come!”
The fact is that for almost everybody the current $20 a month ChatGPT Plus option will be easily sufficient. Perhaps it can do special things (like writing “David Mayer” with no problems), but it seems hard to justify beyond a select few users who need massive computing power. To me, the $200 price point seems more like a price anchor. Essentially, it makes the $20 ChatGPT Plus price point look really good value.
Yesterday, OpenAI kicked off the 12-day event with the announcement that the company’s o1 reasoning model would no longer be in preview, ready for everyone to try.
The AI model thrives with scientific equations and math problems with OpenAI saying o1 can solve 83% of the problems in the International Mathematics Olympiad qualifying exam, a massive improvement on GPT-4o, which only scored 13%. The new model makes fewer errors than the preview version, cutting down on major mistakes by 34%.
That wasn’t the only reveal, however…
Welcome to TechRadar’s “12 days of OpenAI” live blog, where our resident AI experts will be taking you through the next 12 (well, 11) days of everything exciting coming out of the world’s most famous AI company.
What will Sam Altman reveal? How will these new updates and products change the way we use artificial intelligence? Who knows, but we’re incredibly excited to find out.
Services Marketplace – Listings, Bookings & Reviews