Are we living through the Cambrian Explosion of AI?
The increasing volume, some might say avalanche, of new AI models and services being announced on a daily basis might be likened to the Cambrian explosion, the biological ‘big bang’ that happened half a billion years ago and started the evolutionary journey of most modern animals. The launch of ChatGPT AI in November has become an inflection point for AI (ignoring the decades of previous work on AI, including several ‘AI winters’), forcing the major AI players to accelerate the release of their AI research, sometimes in a state not quite ready for prime time ( 谷歌 ’s Bard and Meta ’s Galactica spring to mind). Start-ups abound, with VCs hungrily pursuing AI opportunities like UK-headquarted Stability AI , the power behind the Stable Diffusion AI platform, going from inception to raising $100m in less than 3 years.
The last few days has seen OpenAI release new APIs for ChatGPT and Whisper, a speech-to-text model that is almost certainly going to be used by people to talk to ChatGPT. Meta released LLaMA, a large language model (LLM), 微软 released Kosmos-1, a multimodal LLM, Salesforce partnered with OpenAI to upgrade their Einstein CRM with GPT capabilities, simultaneously injecting AI into the Slack platform. CRM competitor HubSpot launched AI tools to help write marketing content and to manage customer interactions. Privacy-focused search engine DuckDuckGo unveiled a beta-level AI feature called DuckAssist, Midjourney version 5 is with us, allowing users to create even more stunning images from a simple text prompt and Tesla continue to tinker with FSD, their autopilot software, bringing autonomous driving one step closer to reality. And this is just scratching the surface….
This coming week Microsoft is hosting its ‘Future of Work with AI’ (on Thursday 16th) with CEO Satya Nadella and Jared Spataro (Head of Office 365), with rumoured announcements including the release of GPT-4 via their investment in OpenAI, allegedly bringing another multimodal large language model to the world. Multimodel LLMs, like Meta’s LLaMA launched last week, extend AI LLMs beyond just text, to include imagery, audio, video and other forms of data. This allows, for example, users to search, create and manipulate images from a textual description and vice versa. The presence of the head of Office 365 also hints at potential integration of AI functions into everyday office applications – one can imagine writing aids, presentation styles and many other applications. Perhaps the return of ‘clippy’, the Office 97 assistant, “you appear to be writing a procurement contract, would you like me to create a first draft…?”.?
领英推荐
The coming weeks will continue to see incremental updates to existing AIs, new AI platforms, AI added into existing enterprise services, new start-ups chaining AIs in new and imaginative ways and wild new investments in AI. For our AI graduates at the Surrey Institute for People-Centred AI (PAI) , I cannot imagine a more exciting time to start their careers in this fascinating and world-changing technology.
Operations & Systems Integration
1 年Great article - so much happening all at once! I thought this section of the GPT-4 technical paper was interesting: "government innovation policies" as a structural factor that contributes to "acceleration risk" https://cdn.openai.com/papers/gpt-4.pdf#page=57 Certainly a moment which calls for careful consideration.
Professor of Safe AI and Autonomy | Director of Connected and Autonomous Vehicles Lab (CAV-Lab) | Independent Scientific Advisor
1 年Well written and interesting article Andrew! Certainly AI is progressing so fast with huge potentials and of course risks. Russell Stuart discusses AI potentials and risks well in his book ?Human Compatible AI??.