OpenAI CEO Sam Altman has managed to keep the buzz going around the release of a generative pre-trained transformer (GPT)-5 and a ChatGPT search engine by calling previous versions “dumb”. Here’s why enterprises would do well to seek alternatives.
Are we misreading OpenAI’s tea leaves?
OpenAI published a blog in March, hinting at the release of GPT-4.5 Turbo, while Altman referred to a new ChatGPT model this year when asked by a podcaster about GPT-5. On 6 May, Altman posted: “im-a-good-gpt2-chatbot”, adding to the buzz. This model, which can be accessed at “chat.lmsys.org”, has two windows that display responses by competitors (for example, GPT-4 and LlaMA-3) when you ask a question and rate the models ‘good’, ‘bad’, or ‘tie’. The ‘good-gpt2-bot’ outperforms OpenAI’s GPT-4 and Claude Opus on some tasks, suggesting that OpenAI will soon release a new GPT model.
And what about the new search engine?
OpenAI is set to announce “some ChatGPT and GPT-4 updates”. Some believe it will release a ChatGPT search engine, aimed at stealing the thunder from Google whose I/O 2024 event begins on 14 May. Recent reports on Y Combinator’s Hacker News community also suggest that OpenAI has registered a domain name, search.chatgpt.com. Besides, ChatGPT Plus users can already access real-time information using the Bing option. But Altman has queered the pitch by posting that the demo will “…not (be) gpt-5, not a search engine”, but “…some new stuff we think people will love…”
What do these developments imply?
If OpenAI releases a search engine, it will compete with not only Google (the market leader with over 80% share) but also with Microsoft’s Bing that is anyway powered by ChatGPT. But not everyone is happy about Altman’s mixed messages. On 7 May, cognitive scientist Gary Marcus posted on X: “…It’s been 14 months and nothing that merits the name GPT-5 has emerged…”
Why is Microsoft wooing OpenAI rivals?
Microsoft has invested over $10 billion in OpenAI, and uses its GPT models to power Copilot and Bing. This dependence demoralized Microsoft’s in-house AI team, according to Business Insider, prompting it to invest $16 million in French AI startup Mistral AI that rivals OpenAI. Microsoft is also building its own AI model, MAI-1, says The Information. This team is led by Mustafa Suleyman, former CEO of Google’s DeepMind and AI startup Inflection before Microsoft acquired Inflection for $650 million.
Why must firms hedge their LLM bets?
Marcus added in the above-cited post: “… How long before we acknowledge that alternative approaches are needed?” Other than OpenAI’s GPT-3 and -4, enterprises can access Meta’s LlaMA-2 and -3, Anthropic’s Claude 3 model family (Haiku, Sonnet, and Opus), Google’s Gemini 1.0 Ultra and Gemini 1.0 Pro, and large language models (LLMs) from the likes of Mistral, Cohere, and Falcon too. Abacus AI CEO Bindu Reddy said, “…the prudent thing to do is not to be married to any one single LLM.”