Reasoning models
ChatGPT5 isn't coming. Why? It isn't really needed anymore. We have discovered synthetic reasoning. OpenAI was the only game in town for awhile with it's release of the o models — o1 and o3. o1 is available for use and has its chain of thought. How they do this is up for speculation. But it doesn't really matter anymore because we already got an open source model called DeepSeek R-1 and it also reasons. They disclosed in their paper that they use reinforcement learning in order to achieve this. This open source model is as good or better in a lot of the benchmarks. Reasoning models are here to stay and seem to be the path to AGI. Expect many more this year with massive breakthroughs.
Agents
Models are wonderful and have shown us a whole new world. Folks have built incredible agentic products around these models — Cursor, Replit, Lovable to name a few. These products have built specialized agentic solutions on top of the models provided by OpenAI, Anthropic, Google, or built their own. This year we will see a rise in more agent type solutions to new domains. Some domains will seem obvious, some we will wish didn’t happen. These AI companies know that models are one piece of the puzzle when it comes to truly integrating AI into society. Agents are where these models will surface in everyday life.
All of the big players will be releasing official agent solutions this year in some way shape or form. The most obvious example of where this is going is what Anthropic demoed in October with their computer use agent. This year we will see many more agents like this.
AGI breakthroughs
Some folks like to suggest that we are running into limits on what all of this technology can do. With the release of OpenAI’s o3 model and DeepSeek R-1 it’s clear that we’ve unlocked new research we can scale to tackle the problem of AGI. More and more players will be releasing models and research this year backed with benchmarks that show pretty incredible scores measuring their AGI capabilities. I’m so confident significant progress will be made here that discussions around AGI will actually morph into discussions are Artificial Super Intelligence (ASI). That’s how we will end 2025.
China and other countries
If you’ve been consuming any media around AI, you’ve probably heard many of the CEOs of these companies discussing how the United States is ahead of other countries right now and will be for at least the next 5 years. This was due to the United States being responsible for a lot of the technical breakthroughs in research but we also have a chip embargo against China. Generally speaking, Nvidia has made deals with all the big US companies that effectively lock in their chips going to those companies. These GPUs were seen as the key to accelerating the growth needed in AI to get to AGI or ASI.
DeepSeek a Chinese based company effectively said hold my beer with their release of DeepSeek R-1. Showcasing that China is not far behind if at all anymore. They are here and they are open sourcing their models! That’s the promise that OpenAI was supposed to keep but they’ve failed to do so. The US needs to move faster and work even harder if we want to win this race. We should expect China and other countries to be further along but hopefully not ahead by the end of 2025.
Enterprise breakthrough
We are already starting to see this change. Discussing LLMs in enterprise companies outside of AI’s magnificent 7 is becoming common place. The tooling is finally getting to the point where enterprise companies can’t ignore it anymore. This will take over discussions throughout this year. Tie this with agents becoming more mainstream and the demos around them more compelling. There is no question that we will see significant breakthroughs in the enterprise. My prediction is by the end of 2025 a significant company — part of the S&P500 — will declare that they are now utilizing some sort of agent as a replacement for an entire division of their workforce. There will be a significant news story about this at some point this year and this will bubble up to the federal government. Conversations around UBI will increase, ethics, and AI job displacement will be the topic of mainstream media for the foreseeable future. Nobody is safe from these issues.
Let’s connect!
Thanks so much for reading! If you liked this then please let me know down below. I’m also looking to collaborate more with others on Substack. So if that’s you then please reach out to me!
With the news today on Deepseek and how it affected the markets especially NVIDIA, and how certain asks of Deepseek were not answered with questions pertaining to some historical events about China & Taiwan is that possibly a negative for Deepseek? Obviously costs are important to all companies but will that be overlooked to sacrifice factual data for speed and convenience?