Tibo Reveals Google Developed a ChatGPT-Like Model a Year Earlier

What, did Google really develop ChatGPT a year earlier than OpenAI?

It's not that Google backtracked afterward; this time it's an insider from OpenAI—Tibo, head of Codex, just confirmed the news:

That was the previous year's version of ChatGPT. It was originally called LMChat, then later changed to another codename.

Google was too nervous to release it, and DeepMind was prohibited from launching products that could disrupt Google's business.

But why does Tibo know so much about this old story?

The answer is: He was on that Google team at the time.

I was a member of that team at the time.

Yes, Tibo, who now creates "unlimited tokens" for OpenAI, was directly involved in that same project at Google.

Netizens were shocked to learn that Tibo had such an experience:

Nope, I thought you were just a Codex reset button (just kidding).

Here's the thing: a Midjourney engineer recently came across an old interview with Jeff Dean, former head of Google Brain, on X:

Sometimes I think about Jeff Dean’s interview, where he said they had an internal chatbot before ChatGPT, but found it less useful than just using Google Search.

I didn't expect Tibo to personally respond, sparking a virtual conversation with Jeff Dean:

The project Jeff mentioned does exist; it was originally called LMChat and was essentially a year earlier version of ChatGPT.

As for why it wasn’t released, Tibo sighed, saying DeepMind didn’t have the authority—LLMs would disrupt Google Search, and the executives were too nervous to approve it.

After recounting this chapter of the past, Tibo could only let out a sigh:

Sigh, I often think about this.

Netizens were left speechless...

Your former boss at Google should never again be involved in anything related to business.

Similarly, Demis Hassabis, head of Google AI, also said in his biography that OpenAI did not make any groundbreaking innovations. At the time, many people recognized the power of LLMs, but Altman had the courage to disregard safety concerns entirely and released it outright.

But it's not really Google's fault—Google Search started at such a high bar that, if all it does is search, it's already extremely mature.

And what about the large models in 2023? They got almost everything wrong, especially anything involving time-sensitive content—full of hallucinations.

But now, Scaling has solved everything.

“Better than Google Search” — this original benchmark set by Google itself ultimately cost it an unprecedentedly high price.

Tibo, real name Thibault Sottiaux, head of Codex, holder of infinite Token capabilities, aka "Cyber Godfather".

(P.S.: Reset, reset again o(╥﹏╥)o, I accidentally used it all up over the weekend!!!)

In 2015, Tibo joined Google, initially working on Google Maps before moving to DeepMind.

At the time, DeepMind was still a fairly pure research laboratory, with AlphaGo, reinforcement learning, and general artificial intelligence research at the peak of public attention.

At DeepMind, Tibo is primarily responsible for building the AI and machine learning infrastructure used by researchers.

According to Tibo’s recollection, DeepMind internally developed a chat product called LMChat, which later received another codename. Its experience was already very similar to ChatGPT.

But Google’s review was extremely cautious, as it would directly impact existing business, and DeepMind did not have sufficient authority. Ultimately, the product never truly reached the public.

In November 2022, ChatGPT emerged.

When Tibo saw it, he collapsed.

In 2024, he ended his nearly nine-year career at Google and officially joined OpenAI.

It turns out Tibo made the right choice.

The failure of LMChat is not accidental; like Vibe Coding, it’s either zero times or infinitely often (just kidding).

This small story ultimately became a microcosm of Google throughout the AI era.

Today, the lag in Coding's battlefield is sparking an even more intense talent upheaval within Google.

On June 18, Noam Shazeer announced he is joining OpenAI.

This name needs no introduction. One of the core authors of the 2017 Transformer paper, “Attention Is All You Need,” which ignited the AI era. In 2021, he left Google to found Character.AI, and in 2024, Google acquired him and his company for $2.7 billion, making him co-lead of Gemini.

As a result, the $2.7 billion acquisition failed to retain the team for even two years.

On June 19, John Jumper announced his departure to Anthropic.

DeepMind’s vice president, Hassabis, promoted this core lieutenant just six months after earning his doctorate, entrusting him with leadership of the AlphaFold team.

In 2024, he and Hassabis were jointly awarded the Nobel Prize in Chemistry for their groundbreaking work on protein structure prediction with AlphaFold.

But now, the Nobel laureate has chosen a competitor.

After suffering two major blows, Google continues to bleed.

On June 24, two new names appeared on Google's talent bounty list: Jonas Adler and Alexander Pritzel.

Both are core contributors to Gemini. Adler leads AI programming at Google, while Pritzel focuses on pretraining. Both have worked on AlphaFold and are colleagues of John Jumper.

Destination, also Anthropic.

In the comments section of the post, someone asked a very interesting question:

What if Google had released GPT first—how would things be different today?

Unfortunately, not many people responded.

I know how to solve this one!

Isn't this just the classic old joke from the tech world?

In the 1970s, Xerox PARC developed the graphical user interface, the mouse, and Ethernet. These people essentially held all the key cards of the PC era—purely an unbeatable hand.

So what happened? Xerox simply didn’t take this seriously.

Because Xerox’s sales team did the math: Does this flashy interface help me sell two more toner cartridges or two more packs of A4 paper? No. So why bother with it?

...

That’s it—Xerox watched helplessly as Jobs came for a visit, went back, and created the Macintosh, instantly claiming the throne.

From then on, Xerox became Silicon Valley’s scapegoat. Every time the industry underwent a paradigm shift, the big names would drag Xerox out and blame it:

See? This is big company syndrome! (piapiapia, whip sounds)

But honestly, who would be willing to disrupt their own business when sitting in that position, watching billions in printer profits roll in every year...?

Replace "toner and A4 paper" with "search and ads," and replace "graphical interface" with "ChatGPT"—

Google, which wrote the Transformer back in 2017 and held the ultimate gear, ultimately fell behind in the marathon toward AGI...

Just got "America Doubao ".

Reference link: [1] https://x.com/thsottiaux/status/2083596911060324570

This article is from the WeChat public account "Quantum Bit" (ID: QbitAI), authored by Jay.