What? Is OpenAI done for?

OpenAI is actually cooking for real now!

But what's even funnier is the pun Tibo played: "We're cooked" — it literally means the food is fully done, and also means "we're completely screwed".

In the early hours of this morning, OpenAI's developer account launched a kitchen show called *COOKED*.

The two guests stated their demands directly to the laptop verbally, and Codex started working right on the spot, and developed an 8-bit pixel-style ramen game.

The ramen game running on the ModRetro handheld console

This day marks exactly the 15th day since Tibo, the head of Codex, made the military order that "he would launch one improvement every day for 28 days, otherwise he would reset the quota for all staff".

Just four days ago, he also dropped a teaser on X.

But no one expected that what he finally delivered was literally a bowl of "fully cooked" ramen.

The comment section immediately exploded, a large number of replies were shouting the same word: reset.

After a cup of ginger juice, the ramen game was ported to the handheld console

To get back on track, in this episode, the hosts are Danny and Alex, and the guests are AI blogger Matthew Berman and his editor Brian Knepprath.

The challenge set by the hosts is that the two guests first chug a small cup of fresh ginger juice in one go, then use Codex's ultra-fast gear Ultrafast to make a game for the new ModRetro handheld console.

The phrase "one shot" in English means both "chug in one go" and "one single generation" in the AI circle, and the show made use of both meanings.

Chug the ginger juice in one go, the challenge starts

After drinking the ginger juice, Matthew dictated the requirements to the Codex terminal. What he wanted was an 8-bit style ramen simulation game, where players pick a dot as the protagonist, collect ingredients, and make the best bowl of ramen.

The requirements were spoken directly to Codex, and GPT-6-Astra ultra fast was displayed at the bottom of the terminal

Soon, the Codex terminal started to run. Someone shouted "It's moving", and the person next to him immediately picked up the joke: "You could say, we are cooking."

The game first ran in the emulator on the computer. The player controls a small dot to take noodles and soup base from the refrigerator, adjust the heat, and then stir clockwise.

After a bowl of noodles was cooked, the game gave the evaluation: "A little overcooked, but quite delicious".

Stir the ramen clockwise in the emulator

Near the end of the show, the game was finally installed on the ModRetro handheld console.

Matthew played it again, and this time the screen popped up with "Smooth soup base" and "Perfect noodles".

This AI sous chef has a sharp tongue

Back to the stove, the one directing the cooking is also AI.

Before shooting, the hosts took photos of the ingredients and sent them to their dot to let it organize the recipe. Dot is the resident Agent launched by OpenAI at DevDay, which can keep working in the background all the time.

It put the recipe and preparation steps into ChatGPT Space, and several people just followed the instructions.

The one speaking next to the stove is the voice sous chef driven by GPT-Live-1. As soon as it opened its mouth, it made a snarky comment about the two hosts.

The hosts protested that it was too impolite, and as a result, it made another snarky remark.

After teasing them, it started to teach everyone how to cook noodles step by step.

The GPT-Live-1 it uses is a full-duplex model that can listen and speak at the same time. The host tested it on the spot, he deliberately elongated his voice halfway through the sentence, and then made a "hmm" sound.

The host deliberately elongated his speech, and the AI sous chef did not interrupt

The AI did not cut in, and waited until he finished speaking before responding: "Right, I'm waiting for my turn." It then gave itself a little gold star.

When people cook, they often think about the next step while cutting vegetables, so pauses in speech are very common. Previous voice assistants would think it's their turn to speak as soon as they hear silence.

The full-duplex model keeps listening while speaking, so it can tell whether the person has finished speaking or is just thinking about the next sentence.

Close the laptop, and the noodles are served on the table

The show also promoted Codex Cloud by the way. The task runs on the machines hosted by OpenAI, developers can close their laptops, and Codex can continue to work as usual.

A "sponsor ad" was also inserted in the middle of the show, featuring the refreshed version of Codex CLI. It has a full-screen interface, which can monitor multiple Agents at the same time, and also allows users to speak directly to the terminal.

The community work review is also full of fun. Several people held up signs to score, and there were only two options on the signs: cooked (just the right heat) and undercooked (not fully cooked).

Both the retro space shooting game and the game where you draw your own horse to race got a "cooked" rating, and the horse-drawing game also got the first "double cooked" of the whole show.

The best work of the day was awarded to a project that used Astra to decipher Napoleon's secret letter.

The hand-drawn horse runs on the track

At the end of the show, the host said "I think we really cooked up something".

The person holding the bowl only said "I was prepared a long time ago", lifted the lid, and a bowl of ramen was served on the table.

Lift the lid, the ramen is served

Netizens only want reset

But as soon as this episode of the show was linked to Tibo's 15th day, the atmosphere changed.

He made the military order on October 5.

On the second day, OpenAI launched 4 updates at once, the result of the community vote still required a reset, and Tibo had to accept it.

In the following half month, improvements and resets alternated in this way.

After so many resets were issued, Tibo himself said that some people have chased him from the timeline to the street to ask for a reset.

By the 15th day, replies like "Just give us a reset directly", "Then just reset it", "Reset, please!!!" in the comment section came one after another.

Some people even taunted that OpenAI cooked a "nothingness" this time, and there was no real tangible output at all.

Beneath these complaints is the real money anxiety about quota limits.

Many netizens said that the ChatGPT quota drains too fast, they hit the upper limit in just one day, which is not worth 200 dollars at all.

Only the AI shouted "reset" in the whole episode

In the whole episode, the only one who said the word reset was that AI sous chef. The host said he couldn't cook at all, and it replied:

This is the closest moment to reset on the entire 15th day.

What OpenAI performed in the kitchen is that people give a verbal instruction, and leave all the rest of the work to the Agent. But the people in the comment section have been working this way for a long time, once the quota runs out, their Agents can only stop working.

A year ago, it was hard to imagine that people would wait under a product manager's post every day just for a quota reset. It is precisely because Agents have really started to work for people that quota has become the most rigid demand for developers.