Introduction

As we approach the final quarter of 2023, it's worth pausing to consider the likely trajectory of artificial intelligence, particularly in the context of its potential convergence toward Artificial General Intelligence (AGI). This post will attempt to make probabilistic predictions for the technological and social landscape of AI in 2024. Given the rapid advancements in multimodal capabilities, code interpretation in models like GPT-4, we stand at the brink of what might be a seismic shift in AI capabilities.

The Near Future:

Autonomous AI and Corporate Rivalry

GPT-4, released by OpenAI, serves as an important milestone. It has vision capabilities and can interpret code, two significant steps toward higher-level cognitive abilities. What's even more intriguing is the notion that OpenAI experimented with long-term memory [1] in its AI models but hasn't released it officially yet. If all these components are integrated into a unified, autonomous agent, we may be looking at a proto-AGI.

Many seem to have forgotten that German Microsoft CTO leaked details about GPT-4's capabilities, hinting at its potential for video understanding. [2] This raises an interesting question: Is OpenAI deliberately constraining the true capabilities of their technology?

On the other hand, Google has had ample time to study GPT-4 and will release its own exceptional AI model, tentatively called "Gemini." It could either be a leap forward in multimodal understanding or a complete disappointment.

Predictions for 2024

The Arrival of Meta's Llama 3: Meta will release an opensource AI model, Llama 3, with about 90 billion parameters, equivalent to ChatGPT 3.5 and even better at some tasks. This will be a major competitive move against OpenAI's models.
Google Gemini: Will come in several sizes. Most of them a free to use, as they will get incorporated in Google products anyways. There will be a very big one that you can use as a subscriber (like ChatGPT Plus), and the biggest one will be available with a waiting list and not yet opened to the public. The biggest (closed) one will be at the level of OpenAIs 2024 model, while the one you can subscribe to will be around GPT-4 level, but with more capabilities right from the start.
OpenAI's Next Model: In the wake of Google's Gemini (be it a success or a failure), OpenAI will be highly motivated to push the boundaries. Their next m

...

Posts

Wikitag Contributions

Comments