David Sweet (@phinance99) header
David Sweet (@phinance99) profile picture

Bayesian optimizer. Adventure investor. malvin: https://teapawt.net/t.co/2nTfG7tAUu

Manhattan, NY
Joined January 2008
Photo by David Sweet (@phinance99)Photo by David Sweet (@phinance99)Photo by David Sweet (@phinance99)Photo by David Sweet (@phinance99)Photo by David Sweet (@phinance99)Photo by David Sweet (@phinance99)Photo by David Sweet (@phinance99)Photo by David Sweet (@phinance99)Photo by David Sweet (@phinance99)Photo by David Sweet (@phinance99)
And all of the employees who stayed to work with him were not fragile, emotionally-immature fools. And together they created a $4T company. The system worked.
Our latest on Jensen Huang and Nvidia. nytimes.com/2026/11/20/techn…
20
Happy national logging-in day!
1
111
Psst... The arms were the letters. That's what the dance was.
2
43
> And so the point that I’ve made is China is thrilled to have Jensen, who is somebody who was pro-selling into [China].” So Jensen will be back at NVDA once Trump is back from China?
Anduril founder Palmer Luckey says a lot of people are missing the forest for the trees when it comes to the Nvidia saga. He called it the “midwit take,” and said they have it completely backwards. Luckey made the case that most people don’t understand how selling American GPUs into the Chinese market actually works AGAINST the CCP. LUCKEY: “What they don’t understand is China doesn’t want Nvidia to sell GPUs to Chinese companies.” “Like there are Chinese companies that want to buy Nvidia GPUs. There’s 600,000 H-200 GPUs sitting in a port right now, because China will not give them an import permit...” “Because they want to build an internal domestic processor industry, silicon industry, not build one that is dependent on the United States.” “And so the point that I’ve made is China is thrilled to have Jensen, who is somebody who was pro-selling into [China].” “And they’ll say, but Jensen was helping smuggle into China.” “Yes! Against the Chinese government’s will.” “Like the bizarre thing that people don’t get is Jensen was basically on the side of Chinese AI companies, but NOT the Chinese government.”
1
2
52
me: Don't mention the previous implementations in the report. Just describe the current one. claude: This report does not mention previous implementations.
1
2
36
> “People should be angry at me and everyone else developing AI,” Lin said. Well, he's self-flagellating so it's cool. The stuff coming from the Anthropic people is the most ridiculous, immature, socially-inept garbage I've ever heard. (And I've been reading twitter for years.)
BREAKING: An Anthropic researcher says AI will likely leave “pretty much all humans on Earth” permanently unemployable and dependent on government welfare. Tao Lin, who works on safeguards for Anthropic’s Claude models, said in an interview with Palisade Research that he expects “broadly superhuman AI,” as good as or better than the best human expert in almost every field, by around the end of 2027. He said people would eventually “need to strictly rely on government welfare to sustain themselves.” But he doesn’t expect mass unemployment by 2027. He sees a gradual transition in which workers first lose their current roles and move into temporary jobs, partly because “there aren’t enough robots yet.” “People should be angry at me and everyone else developing AI,” Lin said.
2
54
You first.
JUST IN: 🇨🇳 Chinese President Xi Jinping calls to destroy and ban nuclear weapons.
1
13
In a nutshell.
What they want - OAI: capture light cone of all future value. Anthropic: rule world because no-one else is worthy. Google: meet 4th quarter analysts' expectations. ...and that's why I want Google to win.
46
Ontologies are the Yahoo! of the AI era.
19
This is such a narrow view of humanity and society. This is tantamount to saying that humans will no longer want to do anything. Will AI researchers be unemployable? Or will they be so jazzed about AI that he'll want to use it make AI better or understand more or do something new and great? What makes him think everyone else won't feel that way? Sometimes I feel like an anthropologist on twitter. I see the same pattern over and over where the poster imagines that everyone else works differently -- and more simply -- than they do.
Anthropic Researcher Tao Lin: Pretty much all humans on earth will become permanently unemployable because of AI "It's very likely that pretty much all humans on earth will become permanently unemployable" "And will need to strictly rely on government welfare to sustain themselves" "Human employment will go away."
17
Waves of rediscovery as agents roll out to the dev community. - feedback works - standard linters - unit tests - custom linters / "house style". <== HERE - code complexity linter - design reviews by agents - metric feedback - QA testing by agents - formal verification And somewhere along the way: - Multiple agents will make this faster! - Nope. But what if I give the agents different jobs, like in a company! - Nope. But if they could communicate! - Nope. Just run one agent on a project then go set up another agent on a different project. BTW: If you're at the "code complexity" stage, try kiss: github.com/dsweet99/kiss
opinionated linters are booming
3
1
5
131
There is no evidence in any if these events that are clearly cyberattacks. That's worse than when there's evidence.
We break: The highly classified network that the National Security Agency uses for its internal communications is down, according to five people familiar with the matter. The network, called NSA Net, allows employees to communicate classified information internally. At this point there is no evidence of a cyberattack, sources said. As the agency works to understand the scope of the problem, it has enacted staggered arrivals for personnel, the sources said. CNN has asked the National Security Agency for comment. cnn.com/2026/11/17/politics/…
21
David Sweet retweeted
Over the past week, I’ve been overwhelmed by the love and support of friends, industry colleagues, and people I've never met. Thank you to everyone who has shown kindness to my family and me. It means the world to me. I came to the United States 53 years ago, and becoming an American citizen is one of the greatest gifts and proudest moments of my life. NVIDIA has been my life’s work. Building it into the company it is today alongside my colleagues has been a privilege and an honor. Nowhere but in America could a company like NVIDIA be possible. I love our country, and I love NVIDIA. I did not live up to the standards expected of everyone at NVIDIA, and I have made serious mistakes in my personal life. I created problems for the Board that I deeply regret. I own my mistakes completely, and I am committed to cooperating with the Board’s independent investigation. Since my departure, I’ve seen an outpouring of calls from NVIDIA’s employees, investors, business partners, and customers for me to return as CEO. Many continue to reach out to me directly as the company navigates complex technology, market, competitive, and geopolitical dynamics. NVIDIA is a once-in-a-generation company. If the Board of Directors decides that my return is in the best interest of our shareholders, our employees, the company, and America, I would be honored to answer these calls and return as CEO. Thank you again to everyone who has shared their good wishes. I am profoundly humbled by your support. To my family: I love you, and I will spend the rest of my life being worthy of yours. Jensen
61,227
121,752
24,516
1,174,857
59,038,022
"bug" = "bug that allowed the agents to hack us" "no evidence of hacking" = "The agents are very good at this."
30
The best thing you can do right now to add value as a developer or quant is to give this to Opus: "Run QA-style tests until you find a bug." Then go to lunch.
51
We are surrounded by systems too complex to reason about. - The phone in my hand - The internet it's connected to - Any codebase over 10k lines (100k, or something -- no mayter how awesome you are, there's a limit). - Any codebase with third-party dependencies. - Taxes People handle complexity well, both intuitively and through training. The tools are - Build models (mental or explicit) - Focus on one piece at time - Experiment (I would, of course, recommend learning to design good experiments, but I am biased.)
Question to people who use AI to write, review and merge large chunks of their code base: My fear is that this leads to a system too complex to reason about, and therefore we'll end up with production issues that we cannot fix. Can you share how are things looking for you? did complexity explode, or is it well contained? how's troubleshooting production going?
1
71
And we wonder why the young folks aren't having sex.
28
I'm looking forward to their de-slopped model. In the meantime, you can get some improvement from prompting. I use this prompt. It's a list of common problems the agent should find and fix. github.com/dsweet99/malvin/b… - **Mystifying synonymy**: Confusing the reader by referring to the same thing by two different terms. Example: "Dog is man's best friend. You should take good care of your canine." Just say "dog" twice. - **Non-local reference**: An offset (e.g., parenthetical) definition that isn't right next to the noun it defines. Example: "A pound of prevention (an imperial measure of weight) is worth a pound of cure." should be "A pound (an imperial measure of weight) of prevention is worth a pound of cure." - **Misinterpretable "this"** (or "that", "these", "those") are uses of these pronouns where it may be unclear what the antecedent is. It's better to reference it explicitly. Example: "blah blah pretzels. These are making me thirsty." should be "...blah blah pretzels. These pretzels are making me thirsty." - **Unnecessary intro**: It is not necessary to say "X matters because" or "X is important because". We can just tell the reader about the importance directly. - **Unsubstantiated throw-away**: An extra phrase that sounds like "there's more to the story", that you're referring to something well-known that the reader should know, but there's no support in the text through reference or data. Example: "and related settings". You should either say what the related settings are or not mention them. - **Lack of specificity**: It's better to use a specific/precise term than to use a generic one. Example: "The test failed because of a problem with the environment." would be better as "The test failed because MALVIN_CODEX was not set." - **Negative parallelisms**: Contrast framing ("It's not X, it's Y") corrects a claim nobody made, which adds false drama. It's better to just state Y then discuss it. Example: "The optimization didn't improve because of good luck. It improved because we used RAASP sampling." would be better as "The optimization improved because we used RAASP sampling." - **Reflexive rule of three**: The rule of three is a good one, but don't use it if there aren't really three things to communicate. Example: "The new runner is faster, safer, and more reliable" Better: "The new runner finishes in 40 seconds instead of 90." - **Inflated style vocabulary**: These are fancy verbs and adjectives that add tone but no content. (The word "delves" comes up a lot.) Example: "This study delves into the intricate interplay between caching and latency, underscoring its crucial role." Better: "This study measures how caching changes latency." - **List of sentences**: Each sentence in a paragraph is individually clear and relevant but lacks a logical connection to the next. (This often happens during editing, when sentences are polished independently, destroying the paragraph’s natural progression of ideas. ) Transition words won't fix the underlying problem. Try given-new information flow, thematic progression, entity continuation, or rhetorical relation. - **Post-def/No-def**: You use a new or unusual term and either don't define it or define it afterward, leaving the reading wondering what you're talking about. As a rule, define a term (that needs defining) the first time you use it. Example: "We set num_axniks=3 for this study. ... talk about the study ... The parameter num_axniks controls [explain now]."
Why is AI slop so frustrating to read? AI writing often obscures main ideas and invents awkward phrases, making sentences more complex than necessary. Many of these slop patterns violate writing principles established in English writing books. In our benchmark, AI writing breaks these principles far more often than human writing, with Claude Code replies by far the worst offenders. We show examples of the most common slop patterns in this thread
1
110
The models seem to want to do this crappy, inelegant thing where they write different algorithms for different regimes. It's kind of like overfitting. It causes performance cliffs and unpredictable behavior outside of anything that isn't explicitly tested. Of course, they learned this by reading human code, so we have only ourselves to blame.
37