PlunderStruck
233 posts
PlunderStruck
@plunderstruck
AI-native engineer building agentic systems and automation. I ship tools that make teams faster. Opinions are my own.
Katılım Kasım 2020
342 Takip Edilen30 Takipçiler
Most CEOs have no idea what’s really going on
David Senra@davidsenra
IBM built a cloud of suits to make sure the CEO never talked to anyone actually doing the work. @elonmusk does the opposite. "Elon's method is extreme focus on substance. Extreme focus on getting to the truth. In any organization with multiple layers, there's compounding lies. Each layer wants to look good. Each layer puts a little spin on things. If one layer lies to the next layer above it, maybe that's okay. When that happens two or three times, the lies compound. If that happens six times, the lies really compound. If that happens 12 times, the CEO has no idea what's happening. That was IBM. By the time I got there as an intern, I calculated there were 12 layers of management between me and the CEO. They even had a term for it: the great cloud. A cloud of men in gray business suits who followed the CEO around and prevented him from ever talking to anybody who was actually doing the work. When he would come to visit, it was like a visit from the king. A completely impervious bubble. That's the polar opposite of the Elon approach." — @pmarca
English
@BlaseBuffalo @pmarca I knew you would see that and comment on it.
English
Introspection = neuroticism x narcissism x thumbsucking.
English
PlunderStruck retweetledi
Socrates literally said the same thing about books.
We'll be fine.
Sahil@sahill_og
The real danger of AI isn’t job loss. It’s people losing the ability to think.
English
@gdb Too bad my pro usage on codex is gone in 100 prompts if I use 5.4 tho. Would love to experience it more, but I have to use 5.3 to make the usage last
English
hearing this sentiment from many
kache@yacineMTB
5.3 to 5.4 is what i would have expected to warrant a jump to GPT-6
English
@laup30 @petergostev @thsottiaux @romainhuet Yesh something is definitely going on. I tried 5.4 again and my limits barely budged after over an hour of use.
English
@plunderstruck @petergostev @thsottiaux My limits just ran out. Does seem to be chewing through the limits a lot faster than usual. I suspect there is a glitch in the 2x until 2 April, @thsottiaux @romainhuet ??
English
Any Codex niche issues warranting a reset on the horizon @thsottiaux? My Pro account is suffering
English
@KevinNaughtonJr Having worked at Amazon for 4 years, I can confidently say that Amazon is uniquely bad at so many things it's comedic. It's literally a miracle that anything they produce works. Anyone with talent and a will to do a good job simply leaves.
English
and just like that vibe coding is dead
Polymarket@Polymarket
BREAKING: Amazon reportedly holds mandatory meeting after “vibe coded” changes trigger major outages.
English
I find code analysis tools to be quite powerful. I have a shell script that wraps an SCIP index of my codebase that Claude or Codex can use to navigate and verify implementations. I also find that having the agent write itself a checklist plan in an md file, as opposed to using the one in its harness, and telling it to cross it off as it's working keeps it grounded across compactions. When the scope of a project grows very large, it's going to be difficult to avoid compactions.
English
Any task big enough to warrant going to /compact even once is still a task sized too big for LLMs you need to split it out into more phases/parts then.
Also make sure your agents have access to at least Context7 and some sort of search (Brave, Exa, Perplexity all good)
English
I honestly don’t know what people are on about but any work where you let the agent going on and on …
- beyond 25-30 turns
- beyond 50% context fill
- beyond 5-6 spawned sub agents
… is going to come out not just terribly shit, but also you’ll be unable to fix that output
English
English
Search on @Microsoft Outlook sucks. @satyanadella please fix it.
Aakash Gupta@aakashgupta
The fastest way to expose whether a CEO actually uses their own product: make them do the most basic task on camera. Outlook has over 400 million active users. Microsoft’s productivity segment generated $77.8 billion last year. And the official Microsoft support page for “Outlook search not working” tells users to open the Windows Registry Editor and manually create DWORD values. That’s the fix. For a product used by almost every Fortune 500 company on Earth. Edit your registry. The reason Outlook search has been broken for years is the same reason it will stay broken: Microsoft sells to IT procurement, not to the person trying to find last Tuesday’s email. The buyer and the user are completely different people. The CIO signs a 3-year enterprise agreement based on security compliance, Azure integration, and per-seat bundling. Nobody in that purchasing decision opens Outlook and types “Q3 budget” into the search bar to see what happens. This is why Gmail search works and Outlook search doesn’t. Google built for the end user first and sold enterprise later. Microsoft built for the enterprise buyer first and shipped whatever search users would tolerate. 345 million paid seats. The switching cost is so high that Microsoft could ship Outlook with no search at all and most companies would renew anyway. Every CEO of an enterprise software company knows this. The product doesn’t need to be good. It needs to be locked in.
English
@TimTeaFan @thsottiaux My limit just got reset early. What about yours?
English
@plunderstruck @thsottiaux Ok, now it’s almost too late burned through half the weekly limit already 🥲
English
@thsottiaux @vybhavbhadri Yes, and our usage limits are drained! Gimme more t o k e n
English
We have found one issue that leads to some users seeing inconsistent usage across sessions but it is quite rare, affecting less than 1% of users. We are working on a mitigation and continuing the investigation.
For the rest we are not seeing evidence of higher usage consumption other than the advertised token cost increase of GPT-5.4 being 30% higher than GPT 5.2 and GPT-5.3-Codex.
Tibo@thsottiaux
We are investigating reports of higher usage drain than expected for Codex when WebSockets are enabled, the team is investigating and we will provide updates as we go
English
@TimTeaFan @thsottiaux if you go to the codex config.toml add this in:
```
[features]
responses_websockets_v2 = false
```
English
@plunderstruck @thsottiaux Where do I turn websockets off? Either it’s too late here or I‘m stupid, but can’t find it in settings 😅
English
@TimTeaFan @thsottiaux I'm in the same boat. I turned off websockets as soon as I saw this and it seemed to quell the usage rate.
English
@thsottiaux Yeah please look into it, I've burned through 40% of my weekly plan in one day despite the 2x limit until April. That basically never happened. I have a 9-5 and two kids, I can't code this much. This is Opus-level token burn 😅 GPT 5.4 is great btw.
English
PlunderStruck retweetledi
After people get a few months of seeing the impact of AI-induced slop in their codebases, and the resulting slowdown this incurs, I predict we're going to see a mass reversion back to "handwritten" code.
TBH I'm feeling it myself some days. I'm actually faster with Cursor tab and just nailing what needs to happen. AI has a terrible habit of spending time on things that just don't matter.
Might be more "work", but in terms of clock time, I'm getting things done more quickly without involving token inference.
Maybe I'm weird. Maybe this is a terrible prediction (as things frequently are with things that change so quickly), but in this case I've been observing my behavior and others for months and I'm seeing a slow steady trickle back to the "old ways".
English
welcome 5.3 instant!
was proud to help reduce hallucinations
for questions where factuality matters most, it's 26.8% better (when searching) and 19.7% better (when not searching)
English
PlunderStruck retweetledi
we've never done code review
but damn if your team is producing this much code you're using LLMs entirely incorrectly
no one struggles with large amounts of code more than an LLM, if you don't keep that in check you have a self defeating codebase
Latent.Space@latentspacepod
🆕 How to Kill The Code Review latent.space/p/reviews-dead the volume and size of PRs is skyrocketing. @simonw called out StrongDM’s “Dark Factory” last month: no human code, but *also* no human review (!?) in this week’s guest post, @ankitxg makes a 5 step layered playbook for how this can come true.
English
the numbers coming in from this thread:
5090: 166 tok/s (z33b0t), 153 tok/s (EmmanuelMr)
4090: 122 tok/s (StubbyTech)
3090: 112 tok/s (sudo), 100 tok/s (Eduardo)
6800XT: 20-30 tok/s (Dark)
Qwen3.5-35B-A3B. 4-bit quant, 19.7 GB on disk. fits entirely on any single 3090 24GB card with room to spare. no offloading, no splitting, full speed.
5090 owners keep pushing the ceiling and we haven't found it yet. NVIDIA side is stacking up. where are the ROCm numbers? report your GPU and tok/s below. building the full map.
Sudo su@sudoingX
hey if you're running Qwen3.5-35B-A3B on llama.cpp and stuck at 40-70 tok/s on 24GB+ VRAM, you're leaving speed on the table. use llama.cpp from source. add these flags: --cache-type-k q8_0 --cache-type-v q8_0 -np 1 Eduardo went from 50 to 100 on a 3090 (24 GB). StubbyTech just hit 122 on a 4090 (24 GB). full 262K context, zero speed loss. UD-Q4_K_XL quant. all layers on GPU. stop leaving performance on the table.
English
I appreciate your honesty and don't worry about it.
I think our interpretation of the situation might be different though. From my understanding, the DoW wanted to make adjustments to their contract with Anthropic which would allow them to use their AI models for surveillance and autonomous killing (no human in the loop). When Anthropic refused those changes to the contract, the DoW labelled them a supply chain risk (which will cause their company great harm) and accused them of trying to determine the policy of United States Military.
I have no issue with the government terminating contracts and I don't really care that Anthropic lost theirs because they refused to let the military use their models in the way the military wanted. What I have an issue with is the government brow beating, harming, and publicly shitting on an American tech company because they couldn't align on something. The issue is not the contract cancellation, its the labelling them a supply chain risk and telling anyone that does contracts with Anthropic won't be able to do contracts with the DoW. They are an American company that is at the top of their industry. This seems like a reckless power move on the part of the government and something like what the CCP would do. Its just unsettling to see the government behave this way. I don't like it.
English
Ok... I'll apologize for the fascism claim. That was an uncalled for generalization I assumed inappropriately.
Just my opinion, but a military contract is not "a property right" being withdrawn. It's a contract for services, and if the client determines you can no longer meet their specifications, you should expect to lose the contract.
And with respect to information security, concern for your supply chain exposure is a valid complaint.
English
This week, Anthropic delivered a master class in arrogance and betrayal as well as a textbook case of how not to do business with the United States Government or the Pentagon.
Our position has never wavered and will never waver: the Department of War must have full, unrestricted access to Anthropic’s models for every LAWFUL purpose in defense of the Republic.
Instead, @AnthropicAI and its CEO @DarioAmodei, have chosen duplicity. Cloaked in the sanctimonious rhetoric of “effective altruism,” they have attempted to strong-arm the United States military into submission - a cowardly act of corporate virtue-signaling that places Silicon Valley ideology above American lives.
The Terms of Service of Anthropic’s defective altruism will never outweigh the safety, the readiness, or the lives of American troops on the battlefield.
Their true objective is unmistakable: to seize veto power over the operational decisions of the United States military. That is unacceptable.
As President Trump stated on Truth Social, the Commander-in-Chief and the American people alone will determine the destiny of our armed forces, not unelected tech executives.
Anthropic’s stance is fundamentally incompatible with American principles. Their relationship with the United States Armed Forces and the Federal Government has therefore been permanently altered.
In conjunction with the President's directive for the Federal Government to cease all use of Anthropic's technology, I am directing the Department of War to designate Anthropic a Supply-Chain Risk to National Security. Effective immediately, no contractor, supplier, or partner that does business with the United States military may conduct any commercial activity with Anthropic. Anthropic will continue to provide the Department of War its services for a period of no more than six months to allow for a seamless transition to a better and more patriotic service.
America’s warfighters will never be held hostage by the ideological whims of Big Tech. This decision is final.
English