AI in 2024: Predictions

AI in 2024: Predictions

I’ve been thinking a lot about AI in the year ahead. Predicting is tough, but maybe a few of these might come to pass:

  • Google’s Gemini will be better than GPT-4, but not dramatically. And OpenAI will retain an edge in its API and hence retain developer loyalty until its next model release.
  • Beginning with Gemini, we’ll see improvements in model reasoning. New models in 2024 will begin to pass reasoning tests that they previously failed. This will open many new use cases.
  • Other limitations of LLMs will be overcome. For example, knowledge cutoffs will become a thing of the past. Perplexity has already announced models with access to up-to-date knowledge, for example. Hallucinations will also diminish.
  • There will be a greater focus on autonomy. The chat interface will remain, but you’ll see more autonomous “AutoGPT”-style behaviour from AI. We’ll see models begin to execute tasks on users’ behalf, including by using browsers directly, as with MultiOn. OpenAI will increasingly focus here. Relatedly, ChatGPT might get Google Drive and Microsoft OneDrive integration.
  • OpenAI’s GPTs will democratize and expand ChatGPT use, but not make developers rich. People will use ChatGPT for many things they previously didn’t, simply because GPTs make it easy. But developers won’t make huge money. Most GPTs will be easy to copy. Only developers who use proprietary data or APIs will sustain reasonable revenue streams.
  • We’ll see a bunch of new models at GPT-4 level or above, including open source models. This will include Llama 3 and new models from Anthropic and Inflection.
  • OpenAI will release at least one new model, and possibly two. Perhaps a GPT-4.5 in Q2, and GPT-5 around its developer day in November. I think the capabilities included in the release will depend on factors like market readiness and the regulatory environment. OpenAI might hold back on capabilities if the world isn’t ready.
  • ChatGPT will get long-term memory. It will remember things from your conversations, and you’ll be able to edit its memory. (There was a leak about this recently, so it’s less a prediction than an assumption that the leak is true.)
  • Companies will begin strategically prioritizing the use of generative AI. We’ll see this mentioned even more in financial reporting, and executives and managers will begin to have bonuses tied to the use of generative AI.
  • Companies will also begin prioritizing generative AI skills in hiring. As the productivity benefits of AI become ever more apparent, knowing how to leverage generative AI will become an essential skill.
  • The freelance job market will go into steeper decline. Research has already shown that sites like Upwork and Fiverr are seeing reductions in jobs and fees after ChatGPT’s launch. These sites will be the canaries in the coal mine and tracking them will be a way to measure the current and potential future impact of AI on the job market.
  • Instantaneous generation of AI-generated images will open previously unthinkable use cases. For example, imagine walking around in your Vision Pro glasses and having 3D images generated around you in real-time based on what you ask to see.
  • People will begin watching ever-longer AI-generated videos the way they currently watch YouTube and TikTok. Perhaps someone will even launch a video version of a site like Ideogram, where people can create videos from their text. Maybe Netflix buys an AI video generation company.
  • Where’s Apple in all this? Apple will continue laying the groundwork for its long-term AI strategy by developing ever more powerful and more efficient hardware for its devices. It will also announce, possibly at its mid-year developer conference, a future version of Siri that leverages a large language model. But that might only be available in beta by the end of the year. Apple will continue to polish until it has something worth of the Apple brand.
  • The pace of robotics development will increase, and people will become increasingly aware of this. Improvements will include incorporating large multimodal models for things like reasoning and planning. Robots like Digit in Amazon warehouses will begin to show up more, and Tesla will unveil significant advances with Optimus and begin using them in factories.

Okay, those are my thoughts right now.

Please share yours in the comments!

Steffen Jany

PM & Consultant KNO Agency | Content Creation | Tech News

10 个月

I would be very sceptical about GPT-5. There were almost 3 years of development time between GPT-3 and 4 alone and the new model needs more data than ever before for training.

回复

要查看或添加评论,请登录

社区洞察

其他会员也浏览了