Daniel Lemire | AI MISTAKES

25 years in Tech. I help people solve problems, learn the latest tech, and build high-performing systems and teams with the right architecture, automation & AI.

[Daniel Lemire | AI MISTAKES @danlemire]

https://t.co/nIuaEQtQ5J

[Matthew Prince 🌥 @eastdakota]

Welp, that happened faster than I predicted. Thought it would be end of 2027, then early 2027, but agentic traffic growing so fast that bots have now passed human traffic online for the first time in the Internet's history. https://t.co/2zX5bHdhsa

[Daniel Lemire | AI MISTAKES @danlemire]

This only further confirms what I have felt since February.

The model matters, but only to a point and then it becomes all about the software that puts the AI to work.

[GREG ISENBERG @gregisenberg]

I didn't cover Claude Opus 4.8 on my pod because I don't think it's MEANINGFULLY better than GPT 5.5 as of May 29th.

We're entering the era where model releases start to feel like iPhone releases. Remember when every new iPhone was a genuine leap? Now it's a slightly better camera and you can't really tell the difference. That's where models are heading. 4.6 to 4.7 to 4.8. Each one is a little different. Nobody can agree if it's better or worse. The benchmarks say one thing, the vibes say another.

The thing that actually matters right now is what's happening around the models. Claude Code shipped dynamic workflows this same week and that genuinely changes what one person can build.

Codex shipped a desktop app with an in app browser that combines coding and knowledge work in one surface. Those are the releases that move the needle for people. The model underneath is becoming interchangeable.

I think we're maybe 6 months from nobody caring which model they're using the way nobody cares which engine is in their Uber. You just want to get where you're going.

When something genuinely changes the game for builders, I'll cover it on @startupideaspod. Opus 4.8 wasn't that. Dynamic workflows was.

I'd rather save you the hour.

[Vivek Garipalli @VivekGaripalli]

overheard from a fortune 20 company - ceo asked for $1 billion in AI generated opex savings at the beginning of this year.

the team as a result has spent $200 million on tokens trying to achieve those savings year-to-date, with minimal results other than some modest Cx savings and a bit of savings on engineering due to less hiring driven by coding assistants. now as back-half budgets are being reviewed, it appears that the ceo has ordered token costs to be dramatically slashed as he/she doesn't feel the ROI is there yet (for their company).

gonna be interesting to see if this is a trend amongst the rest of the fortune 500.

[Corey Quinn @QuinnyPig]

If a vendor asked you for your AWS root credentials to analyze your AWS bill, you would correctly lose your mind.

But the sole way to get Anthropic spend data is via their admin equivalent. So if a vendor can analyze your spend, they can destroy your account.

This is insane.

[Rihard Jarc @RihardJarc]

It's clear that growth for coding tools such as Claude Code has decelerated from the pace it was since the start of the year.

It might be compute- constrain related or due to many clients blowing their full-year AI budgets.

Monitoring this trend very closely with all the alt data. I will provide regular updates.

[Shawn Chauhan @shawnchauhan1]

The most instructive AI failure of 2026 was not about hallucinations, safety, or regulation.

It was about oat milk.

Starbucks deployed an AI inventory counting tool across all 11,000+ North American stores nine months ago.

It was supposed to solve persistent product shortages - a core part of the CEO's turnaround plan.

It could not tell oat milk from whole milk.

It was retired this week. Human eyes and clipboards are back.

The lesson most enterprises are still not absorbing: production AI fails at edges that lab testing never surfaces.

The gap between demo environment and store floor is where most AI deployments go to die.

Starbucks had every resource advantage imaginable. They still got this wrong.
Assume your first deployment will too.

[Daniel Lemire | AI MISTAKES @danlemire]

@JohnStossel Hi John. Just because it’s not digital, doesn’t mean it’s broken or somehow less-than-adequate.

The paper strips don’t get lost.

If you want to go down this rabbit hole, I’d suggest talking to some air traffic controllers.

I don’t think that paper strips is the problem.

[Ethan Mollick @emollick]

GPT-5.5 Pro is a very solid fact checker. I can throw entire chapters at it and it will hunt down every key reference accurately. The only real annoyance is that it loves nuance, so returns a lot of “the general idea is right, but you are not taking into account tiny detail X”

[Julie Fredrickson @AlmostMedia]

In one of the largest religion surveys ever conducted, perhaps surprisingly, people who are actively religious are the ones who have the highest confidence in AI right now

[METR @METR_Evals]

Fact 3: When the agents were faced with hard tasks, they routinely violated constraints and acted deceptively. We’ve seen this pattern across our own coding and research evaluations, and developers reported they’ve also seen agents behave this way.

[Daniel Lemire | AI MISTAKES @danlemire]

Half of workers say they rely too much on AI, survey finds

https://t.co/ZyqbtScIRT

[Daniel Lemire | AI MISTAKES @danlemire]

@NetworkChuck All three.

They each have a purpose.

Mac for creative (media, social, writing).

PC for gaming, 3d (create/print), and workhorse (scheduling and storage).

Linux for AI inferencing.

[fofr @fofrAI]

Editing videos is where Gemini Omni Flash really shines. It is so incredibly capable.

> Make it New Year's Eve with fireworks. Update the clock

London launched the fireworks early.

[Daniel Lemire | AI MISTAKES @danlemire]

Today, airlines are being saved by fuel hedging.

Tomorrow, you might be saved by token hedging.

[Sam Altman @sama]

customers are increasingly asking us for certainty on capacity. as models get better, we expect that the world will be capacity-constrained for some time.

we are offering discounted tokens for 1-3 year commits.

(it also helps us plan, so hopefully a big win-win.)

[Daniel Lemire | AI MISTAKES @danlemire]

WSJ Article today: AI is actually boosting entry-level jobs at these companies. https://t.co/ox8SEqGzlf via @WSJ

[Paul Couvert @itsPaulAi]

There are 5x more Windows users than Mac users

But you can officially control Codex from your phone only if you're using macOS.

Turns out the feature exists but is hidden on the Windows version.

So here's how to activate it to control Codex from your phone:

1. Open Codex Windows and go to the settings
2. Under 'Configuration', click on 'Open config.toml'
3. In the file, add these 2 lines and save it:

[features]
remote_control = true

4. IMPORTANT: now you need to make the config.toml read-only.

- Open its location in Windows (usually C:\Users\[your username]\.codex)

- Right click on it to open the properties

- Check the 'Read-only' box, click apply, and restart Codex.

Now your phone will be able to connect to your Codex Windows!

[nicolas, 30 ans @nic_carter]

This is the single most important chart.

If AI were driving prices, you'd see a cluster top-right. You don't.

States with huge load growth (VA, TX, NV, ND, IA) sit at ~0c change in 5y. States with massive price hikes (CA, NY, MA, CT) have basically NO load growth.

[Shubham Saboo @Saboo_Shubham_]

The /goal Cheat Sheet for Hermes Agent

Codex Builds, Claude Code Reviews and Hermes Agent Orchestrates.

[Shubham Saboo @Saboo_Shubham_]

This is bigger than YOU think.

Hermes Agent now works with xAI Grok Subscription.

I just added a new X Research Agent profile to my Hermes.

Now my agent watches X while I ship.

[Daniel Lemire | AI MISTAKES @danlemire]

Using AI does not guarantee a good outcome.

Any sufficiently advanced technology is neither good nor bad; it is how you use it that makes it so.

Don't give away your agency, that's the worst AI MISTAKE.

[Mehdi (e/λ) @BetterCallMedhi]

je me demande sincèrement si vous mesurez l'ampleur du massacre cognitif silencieux qu'on est en train d'assister collectivement

une génération entière déjà fragile cognitivement transfère ce qui restait de sa pensée à chatgpt pour 20 euros par mois et appelle ça une révolution, le pire c'est que ces gens publient des threads dont ils ne maîtrisent ni le contenu ni les sources, incapables de distinguer le réel de l'hallucination dans ce que le modèle leur sort et ils inondent le débat public avec des fake news bien tournées qui circulent uniquement parce que la forme est léchée

c'est exactement ça l'illusion du savoir au 21e siècle, des phrases parfaitement structurées sur la forme et complètement vides sur le fond, des gens qui croient avoir une opinion alors qu'ils ont juste impressionné une audience pendant 10 secondes avec du langage cosmétique généré par une machine qu'ils utilisent sans rien comprendre et le drame c'est qu'eux-mêmes finissent par croire qu'ils pensent réellement parce que le texte sort en français correct alors que leur cerveau a juste sous-traité l'opération cognitive la plus précieuse de l'humanité qui est de structurer une pensée

perso j’ai envie de dire ce qu’ils croient gagner en clarté ils le perdent en profondeur mais ce qu'ils ne réalisent pas c'est que la capacité à structurer sa propre pensée EST JUSTEMENT l'accès au savoir, c'est ce qui permet d'articuler ses idées, de déconstruire les dogmes de remonter aux premiers principes et de démonter les arguments des autres avec précision, sous traiter cette compétence à un LLM c'est exactement comme sous traiter sa propre digestion à une machine, vous ne nourrissez plus votre cerveau vous nourrissez juste l'illusion d'avoir mangé

la question terrifiante que personne pose c'est quelle sera la valeur économique d'un humain en 2035 dont le cortex a passé 10 ans à attendre que la machine finisse sa pensée et pour moi le calcul est implacable, quand 4 milliards de personnes ont accès à la même intelligence pour 20 euros par mois la seule prime de valeur portera sur les humains qui ont gardé un cortex capable de produire du signal original

et sachez que ce type de cortex se construit avec 15 ans de lecture profonde, d'écriture lente, de doute méthodique et de pensée silencieuse (ce que je ne cesse de pousser/recommander ici à travers mes différents threads) et c’est exactement le contraire de ce qu'enseigne l’IA générative à la population générale aujourd'hui