I watch far too much YouTube for my own good. You have to sift through numerous ill-informed and overly sensationalistic titles and articles to find a few individuals who conduct actual research and provide somewhat informative content. As I discover these and when they have something interesting to say, I thought I’d present them to our audience. If it’s successful and interesting, I’ll continue to do it. But in the light of New Year’s predictions, this YouTuber has predictions backed by evidence, and they are also consistent with what I have been finding.
According to this video, in 2024, OpenAI’s ChatGPT is poised to take a significant leap forward. Recently, Sam Altman, OpenAI’s CEO, sparked excitement with a Twitter thread detailing the roadmap for ChatGPT. Key areas of focus include enhanced reasoning, the much-anticipated GPT-5, improved voice modes, and a revolutionary feature termed ChatGPT with Memory.
Improved reasoning is a top priority for OpenAI. ChatGPT is excellent at writing text and has made vast improvements on its math abilities. But that disguises the fact that it struggles with true reasoning. What the world mistakenly labels as ‘hallucinations’ are actually due to ChatGPT not knowing what it doesn’t know or when it’s wrong. It lacks the basic reasoning skills that humans have.
There are innovations in prompting like Chain of Thought which compensate for the lack of more sophisticated reasoning capabilities. However, there have been developments in other open-source models that demonstrate much more nuanced and accurate responses.
GPT-5 is officially in training, fueling speculation about its potential superiority over GPT-4. With competitors like Google’s Gemini making waves, OpenAI is going to be highly motivated to maintain its edge. The trademark for GPT-5 suggests groundbreaking advancements, particularly in image capabilities and possibly video.
Perhaps the most exciting development is ChatGPT with Memory, known internally as Project Sunshine. This feature enables ChatGPT to learn from previous interactions, making it more personalized and efficient.
Imagine if ChatGPT could actually remember all of your conversations and learn about you. This continuous learning approach could revolutionize how users interact with AI, offering a more tailored and context-aware experience.
The downside? ChatGPT is going to have to get more secure. Another video I’ll show you this week will illustrate how easy it is to bust through ChatGPT’s guardrails and obtain information that you should not be able to get this easily. Until that is solved, implementing a memory of all your interactions is a security nightmare.
Voice interaction is set to become more seamless and intuitive. The integration of Whisper’s API for speech recognition points to a future where users can engage with ChatGPT in their native language, hands-free, and with high accuracy.
One barrier to developing applications using ChatGPT’s API is how do you control and pass on the costs? While many calls to the API cost only pennies, it’s not difficult to imagine scenarios where the API costs could skyrocket.
But what if you could have a login that maintained the costs on the user’s account and API key? That would allow the creation and easy distribution of applications leveraging OpenAI’s API. This could be a game-changer in the development of sophisticated applications using ChatGPT – particularly when enhanced reasoning and other features from GPT-5 are available.
These are some of the key predictions covered in this video by the YouTuber who refers to his program as AIGRID. As I noted earlier, his work tends to be fairly well researched. I’d put my money on these predictions for 2024.
