AI in 2023 through 10 Images
“In 2024, Be Prepared for Another Year of Exponential Growth” — Kevin Scott CTO and executive vice president of AI at Microsoft.
#1 Massive Investments in AI
In the first half of 2023, investment in AI surpassed the total funding of the previous four years!
The Top companies that have raised this year’s largest equity rounds are:
- OpenAI with a $10B investment from Microsoft
- Inflection, which focuses on human-computer interfaces with $1.3B Series B
- Anthropic, an AI model developer and research outfit with $850M across a Series C! And this does NOT account for Anthropic rounds from Google and Amazon with respectively $2B and $4B investments!
- Adept, developer of an enterprise-focused AI model called ACT-1 ($350M Series B)
- Cohere, AI model developer for text generation, classification tasks, and search ($270M Series C)
- Mistral AI surprising everyone with two successive rounds of $100M early 2023 then $400M in December 2023!
#2 Detailed Funding Recap— Jan to Dec 2023
#3 Massive Growth and Revenue — Big Winners?
- OpenAI went from $0 to $1 billion in revenue in one year!
- Nvidia’s stock grew more than threefold in 2023, entering the realm of trillion-dollar companies!
- Midjourney, the renowned bootstrapped AI company, is on course to announce $300 million in revenue with fewer than 100 employees and has raised $0 in funds so far!
#4 Proprietary Models vs Open Source — Catching Up
As of today, December 31st 2023, Mistral 8x7B is the best Open Source model on the Elo Rating. And it beats GPT3.5!
Note: Elo Rating is a method for evaluating models in a blind, head-to-head format. This process has been crowdsourced and has accumulated over 130,000 evaluations, providing a subjective assessment of the quality of the outputs from different models.
#5 Most Impactful AI Releases
It would be impossible to cover everything that happened in 2023! But here’s a reasonable recap!
Notably, we can mention,
- LLaMA then LLaMA2: Open Source Models Fueling Innovation. Often Used as the Base Model for a Wide Spectrum of Finely-Tuned Open Source Models.
- GPT-4Vision: capable of interacting with both text and images.
- Gemini: the rise of multimodl models. Still in early stages ; 2024 will be the year of multimodal models (managing text, images, videos and audio all at once!).
Everything your need to know about GPT-4Vision and Gemini family of models.
- Phi1, Phi1.5 then Phi2: Compact models, like Phi2 with 2.7 billion parameters, perform as well as models that are 25 times larger.
- Mistral 8x7B: as mentioned earlier, this is the first model beating GPT3.5 on Elo Ratings (which means people prefer its output to gpt3.5's)!
- Midjourney, Dalle, Stable Diffusion: highly performing image generation models (which one is the best? check this here)
#6 AI Powered Developers Win!
GitHub Copilot, announced that it generates 46% of developers’ code on average, with a 90% increase in task completion speed!
If you are not leveraging AI, you are already falling behind!
#7 AI Video Generation is accelerating!
In January, there were no public text-to-video models. Now, there are dozens of video gen products and millions of users (source)
#8 AI Impact of Businesses — the (not so) curious case of Stackoverflow
StackOverflow has laid off 28% of its staff as generative AI for coders gains momentum!
StackOverflow, long considered a coder’s best friend, has been losing momentum as an increasing number of developers turn to generative AI tools such as ChatGPT, GitHub Copilot, and various open-source models to write, update, and fix code. With more than 56% of its traffic lost, StackOverflow announces laying off 28% of its staff (~100 employees)!
This is one notorious case of several businesses already disrupted by the progress made by Gen AI!
#9 Medical AI matches best Doctors!
The Med-PaLM 2 model developed by Deepmind has reached a remarkable accuracy rate of 86.5%, which is comparable to that of experts!
This model facilitates interaction between clinicians and the AI, allowing them to engage in detailed discussions about intricate medical conditions. Healthcare organizations can already access this technology via Google Cloud’s MedLM.
Additionally, a multimodal version of Med-PaLM was demonstrated, exhibiting its ability to analyze and interpret various types of data, including medical images and text. This development illustrates what’s coming next in this field!
#10 Towards General Purpose Robotics
Imagine a Multimodal model embedded in a robot.
RTX2: the ChatGPT of robotic models. Google Deepmind proved that a model trained on a large variety of general purpose videos (task oriented) outperforms models trained on a specific task.
Now imagine such a model running on Optimus Gen2 (by Tesla) or Boston Dynamics robots!
Conclusion
What a year!
what about you? What excited you in 2023? And what are you expecting in 2024?
Happy New Year!