TRANSCRIPT
Hey, it’s Jim Love, your host of Hashtag Trending. Welcome to another edition of Holiday Bytes. There’s been a lot of chatter about what’s happening at OpenAI, driven by some postings on the OpenAI site and some items that got posted and then disappeared.
Sam Altman, OpenAI’s CEO, has shared a roadmap featuring enhanced reasoning, the debut of GPT 5, improved voice models, and some exciting new features including ChatGPT with memory.
Now, based on all of that, here are some predictions for OpenAI in the coming year.
These predictions, while focused on chat GPT, highlight the fierce competition in AI. With players like Anthropic’s ClodeAI, Google’s Gemini, and emerging open-source alternatives like Mistral AI, the race is heating up. Altman’s roadmap, though groundbreaking, is just the entry ticket to this new competition.
Enhanced reasoning, a new frontier.
ChatGPT’s eloquence masks a key limitation. It lacks real reasoning. It knows the likely next word, it seems very real, but it doesn’t truly understand what it’s talking about. It doesn’t know truth from fiction. It can’t discern right from wrong. It doesn’t know what it doesn’t know.
And what we see is hallucinations are often just ChatGPT veering off course, confidently continuing with the most probable text, regardless of accuracy. Now there are workarounds. Fine-tuning and prompt engineering can help. They mitigate some of the limitations. Techniques like “Chain of Thought” can correct issues and enhance accuracy.
And I’m going to delve into some super prompts later this week for those interested.
You can make it more accurate, but these are just temporary fixes. Recent developments in open-source models are showing promise in logical reasoning. I’ve seen a demo where an open-source model seems to genuinely understand and logically process its responses, and crack some logical problems that would defeat open AI’s 4.5 version.
Now, OpenAI is pushing these boundaries, aiming for a more nuanced and accurate chat GPT. But remember, Altman’s ultimate goal with OpenAI is artificial general intelligence. In a recent interview, he expressed confidence that a large language model could achieve AGI sooner than we think.
GPT 5 is in training.
Expected to outshine chat GPT 4. Altman hinted at this before his departure last year. Mira Murati, now the chief technology officer, had to step in to prevent him from giving away too much in that interview. Now with rivals like Google’s Gemini, open AI is under pressure to stay ahead, especially with the potential enhancements in its image and video processing and logical reasoning.
And don’t let that video reference slide by. Video is going to play a huge role in generative AI going forward. What it can do already from a number of offerings is amazing. It’s a great editing tool right now, and it speeds up production, but it’s going to get better and better in terms of the creation of video.
And there will be synergy between this analysis and production. The more generative AI studies video, the better it’s going to be at producing it.
But video is also an enormous source of unstructured and extremely relevant content, Video can convey so much more information than text going beyond facts to sentiment to emotion and more.
So look to video, not just in terms of creation or editing, but as a new content source for training and what that can do as we move forward towards making AI have an even greater understanding of our emotional as well as our factual needs and behaviors.
ChatGPT with memory
One of ChatGPT’s current limitations is its inability to remember past interactions, even in the same conversation sometimes. This is partially addressed through stored user instructions, but that’s just a temporary fix.
Imagine if ChatGPT could recall all its interactions with you. The possibilities for personalization are now immense. OpenAI is reportedly working on this. Project Sunshine or ChatGPT with memory is what it’s called. This could revolutionize user experience and make it much more personalized and efficient.
However, a significant hurdle to this is – security. I’ve seen demonstrations of how easily chat GPT security can be breached and how prompts can be revealed. So before this memory feature goes live, expect major security enhancements.
Voice interaction: Beyond the Turing Test.
Yeah. If you’ve only used Siri, Alexa or Google for voice commands, prepare to be impressed by OpenAI’s capabilities. You can check it out now.
Their AI can engage in almost human-like conversations.
I’ve actually done an interview with the AI at a conference that I was hosting, and the only giveaway was a slight delay in response, something Google’s Gemini cleverly omitted in its demo.
And this year, voice interactions with AI become even more natural. The integration of the Whisper API suggests a future of hands-free, accurate interactions, and in every native language.
Sign in with OpenAI
Applications using ChatGPT’s API currently involve some cost considerations. Each API call, though inexpensive, can add up, especially with complex calculations or with large user bases. So the proposal for OpenAI’s new login system where costs are tied to the actual end user’s account instead of the developer’s account could be a game changer for developing sophisticated applications
And the app store
Despite being delayed by Altman’s firing and rehiring, the concept of an AI app store remains compelling. Apple reportedly makes 80 billion a year or more from their app store. And if you saw Altman’s developer conference launch, you’ll know, he’s obviously been inspired by Steve Jobs, not just in presentation style, but he studied Jobs’ strategy.
And with OpenAI needing funds for ongoing operations and ambitious projects, I think this idea of the App Store is far from dead. It’s not the sort of genie you could put back in the bottle.
And those are my predictions for 2024 and that’s it for this holiday bite.
And if you want to see more from YouTube, check out the Best of YouTube series that I’m launching this week. I’d love to hear your opinion. I’d love to know whether we should continue it.
We’ll be back with our interview show on the weekend of January 6th and our daily show resumes on January 8th. Until then, wishing you all a fantastic new year.