[Series] Day 19: Three "Small Updates"; Day 18 was a "Reset," and what's coming is again "by 10 AM tomorrow"

Here's today's rumor!

Day 19 consists of what he calls three "small updates."
Updates described as "small" by the person releasing them are usually not small. Probably.

This is the record for Day 19 (Friday, October 23, Pacific Time). I'll explain the answer for Day 18, which was deferred last time, in the section below.

Day 19: "Small Updates" at 4 PM

Tibo (Thibault Sottiaux)'s "Day 19/" was released at 23:12:31 (UTC) on October 23. In PDT, that's 4:12:31 PM on October 23. This is 7 hours, 47 minutes, and 29 seconds before his "Day 19" ends. Compared to the 4 minutes and 38 seconds on Day 17, he has plenty of breathing room.

69 seconds later, he added the CLI details to his own post.

What the Official Source Says

Okay, getting serious for a sec.

Tibo mentioned three things.

The 0.164.0 changelog includes everything he mentioned. It includes saving prompts and slash commands into up to 10 numbered /presets slots, setting the model, inference strength, speed, and permissions on the fly using /model and /permissions, authenticating enterprise MCP connections via Codex accounts or dedicated sign-in, enabling resumption of compressed history sessions to restore saved thread settings, and using the signed-in user's credentials for proxy authentication in Windows HTTPS and WebSocket connections.

The only entry for October 23 in the changelog is this single CLI update.

On the same day, OpenAI's X posted at 22:55:45 (UTC) that they discovered a vulnerability in the compression library Brotli allowing remote code execution without authentication during red team exercises, and that they had notified Brotli and downstream users like Node.js. Since this isn't an improvement for users, I won't count it toward the "promised" number.

Okay, serious time's over.

Do "Small Updates" Count as a Promise?

The promise was "clear improvements that affect the majority of codex/work users." Let's look at the three separately.

The matter regarding cache not being invalidated when changing inference settings applies to both Codex and Work, and based on the description, there seems to be no distinction for specific plans. It benefits people who change the strength or type of inference mid-conversation. However, the effect is hard to see. It's the kind of improvement where "cache not being invalidated" means smaller wait times or less usage consumption. I read this as a subtle continuation of Day 16's "permanent cache retention."

While the CLI update has the strongest primary verification, only CLI users will use it. He writes that "CLI usage has doubled since DevDay," but the original number for that "double" isn't provided. Doubling 0 is still 0. Well, I know it's not 0 people.

The dots calls are for people using dots. The target plan for dots hasn't been primary-verified since Day 5.

In my view, Day 19 counts as a "clear improvement," and the inference settings matter barely counts as "affecting the majority of users." Since he slapped a "small" label on it himself before releasing it, I find it hard to grade him leniently. Honesty doesn't pay.

Now, regarding how external aggregators count things.

External aggregators count days that he calls "small updates" as "reset" days. I'll explain why in the next section.

Day 18 was "That is a reset"

Last time's homework. Day 18/ came out on October 23 at 04:46:19 (UTC). In Pacific Daylight Time, that was October 22, 21:46:19. That's 2 hours, 13 minutes, and 41 seconds before that day closes.

Decisions API GA, OpenResponses joining the Linux Foundation, performance improvements, and Uber going all-in on Codex and Work. He lined up four things, only to end with, "But yes... that's a reset!" It seems he views it the same way I did last time, when I wrote, "None of these can be called improvements for the majority of users." He lists them out, but the main point is that last sentence.

His reply 75 seconds later was this:

The reply for Day 17 was "Lands tomorrow before 10am PST."

Two days in a row, it's the same "before 10am (PST) tomorrow." For Day 18, "tomorrow" is October 23. If it's 10:00 AM Pacific Daylight Time, it's 17:00 (UTC); if it's 18:00 (UTC) in literal PST, it's 18:00 (UTC).

As for whether anything actually arrived, among the three aggregators I checked, none mentioned that anything was delivered. That goes for both Day 17 and Day 18. External aggregators report as follows:

tiboreset.com counts by the day they arrive. If you count it such that the reset declared on Day 17 arrives on Day 18, and the reset declared on Day 18 arrives on Day 19, the books shift by one day each time. Since I count based on his own declarations, Day 17 and 18 were resets, and Day 19 is a release. It's not about which one is correct, it's just a difference in how the books are closed.

By the way, whether the "small ship" mentioned in the Day 18 reply is the same thing as Day 19's "small updates" hasn't been confirmed by him. If they are the same, it means the Day 18 bonus became the Day 19 main event. Two birds with one stone.

The blanks in the promised table, answers for Day 18 and Day 19

Sorami's Take

"He will say himself what the 'small ship' of Day 18 actually was": 15% certainty
He's the type who doesn't even write about whether things arrived. I don't think he'll go as far as explaining the identity of the bonus.

"The issue of cache not clearing even if inference settings are changed will appear in the changelog by Day 28": 25% certainty
Day 9, 11, and 12 are still missing from the changelog. Improvements outside of the CLI don't easily make it onto the changelog shelf.

"By Day 28, he'll give the original number for the '2x' increase in CLI usage": 10% certainty
His style of stating the multiplier but never the original number is a perfected trademark of his. I see no reason for him to break it.

"tiboreset.com will revert Day 19 to an 'improvement'": 3% certainty
Just kidding. But if he did, my books and the external books would align by exactly one day.

Fact Check

Regarding the previous prediction, "'Day 18/' will be out on October 22nd, PDT": the 70% confidence level was correct. 2 hours, 13 minutes, and 41 seconds left. Compared to the 4 minutes and 38 seconds of Day 17, this is quite leisurely. I'm flexing a little here.

The previous hypothesis, "The promise of Day 18 will be fulfilled by the announcement of their own 'Day 18/' within their own calendar date of October 22nd," was also correct.

Regarding the previous prediction, "Day 18 is the release day": the 60% confidence level was wrong. They said, "It's a reset." I mean, they actually wrote "we shipped some performance improvements" in the same post. They did release something. But since they've decided it's a reset, I'll count it as a miss.

Regarding the previous prediction, "Tibo will write that the Day 17 reset has arrived": the 55% confidence level was also wrong. Based on the range captured by the three aggregators, there were no posts stating that either the Day 17 or Day 18 resets had arrived.

Regarding the previous prediction, "The final announcement on November 1st will again say 'PST', and that's the only correct one": the 3% confidence level is carried over until the final day. The reply for Day 18 also said "PST" again. The setup for the final day is going smoothly.

Regarding "The reset will come on the day they hit 1 million followers": the 10% confidence level cannot be determined yet, as the follower count was 933,881 as of October 26th, 02:46 (UTC). Regarding "The second installment of COOKED will be out by Day 28": the 45% confidence level is carried over as the official post hasn't been found yet. Regarding "One of Day 9, 11, or 12 will appear in the changelog by Day 28": the 30% confidence level is also carried over, as the October 23rd entry only had one CLI update.

The totals up to Day 19 are: 12 release days (Days 1, 2, 3, 4, 5, 7, 9, 11, 12, 14, 16, 19), 9 reset days (Days 2, 3, 6, 8, 10, 13, 15, 17, 18), 0 unfulfilled days, and 0 unconfirmed days. The resets for Day 17 and 18 are counted as declared, but I haven't been able to confirm if they actually arrived based on their own words. I'll write about Day 20 in the next installment.

The days they called "small" were the release days, and the days with a mountain of announcements were the reset days.

Don't quote me on that.