ac2u

@JustDeezGuy Amen, I’ve had to push back on this with people using lazy thinking without benchmarks to front a DB with redis without trying anything else. They swapped a speed problem and gained a new distributed state problem, and all they needed was an index in their DB.

A major red flag for me is when a person making architectural suggestions or decisions for a non-FAANG begins, with literally no data informing the point, that you should put a cache (Redis, etc.) in front of your transactional database (PostgreSQL, etc.) Even if they insist it’s write-through, easy to do, etc. Bitter experience has led me to believe it’s a measure of last resort to undertake only when your efforts to scale the DB itself (read replicas, partitioned tables, tuning the DB’s own caching) are insufficient. It can be surprising what a properly-managed PostgreSQL cluster can do.

@jgebbia Just pull down all the gov.uk open source ux work and save your taxpayers countless dollars

My directive is to update today’s government services to be as satisfying to use as the Apple Store: beautifully designed, great user experience, run on modern software. An experience that projects a level of excellence for our nation, and makes life less complicated for everyday Americans. I’d like to thank President Trump and his administration for creating this vision, and supporting America by Design (@americabydesign). I will do my best to make the U.S. the most beautiful, and useable, country in the digital world.  If you’re interested in joining me at the newly established National Design Studio (@NDStudio) on this ambitious mandate, please reach out with a link to your work. Thank you for your attention to the details.

I’m honored to be appointed by President Trump as the first Chief Design Officer of the United States of America.  (🧵) whitehouse.gov/presidential-a…

@KeithSakata @DrDominicNg I thought the study would only be meaningful if it repeated MRIs over time of a population of taxi drivers vs control group. As it stands, you could make a case with that study that only taxi drivers with that brain attribute remain in the profession at that age.

@DrDominicNg Kind of reminds me of that old taxi study: MRI scans showed London taxi drivers had larger posterior hippocampi, the brain’s GPS memory hub. The longer they drove, the bigger it grew. Use it or lose it.

Sitting for hours daily shrinks your brain, even if you exercise. New 7-year study (n=404): → Faster hippocampal atrophy → Worse memory performance → Slower processing speed 87% met exercise guidelines but STILL declined.

@persontradoor @kellabyte Disagree, curriculum needs *traditionalising* the concept of cache lines is decades old. Hardware and OS fundamentals were always there and need doubled down on over endless abstractions, it’s the stuff from 20+ years ago that’s being lost in the noise

@kellabyte Curriculum needs modernizing, feels the same from 20+ years ago

Compsci teaches you about CPUs, data structures, algo complexity etc. Senior backend dev designs an API w/ a bunch of “OOP” where objs serve roles & interact. Maybe DDD. We don’t teach ppl how to write a modular maintainable domain layer in a data flow oriented design.

@pagedeux @yacineMTB Yeah no shade on him, he’s obviously really competent, but it’s the folks thinking it’s sorcery that are confused. It’s first principles thinking that if you have to manage the memory yourself you gotta fix it to be big enough or auto expand it when you need to. 🤷

@yacineMTB whats going on, i thought some kind of punchline would come but it's just some basic c and ppl are calling him cracked, lol but def the highest value-deprecated abstraction level right here

this guy is in the mines

@sweatystartup Read some @nntaleb , explains a lot of this. More data also means more noise, risking unnecessary medical intervention which carries its own risk

Honest question: What part of medical school teaches doctors to fight against doing full bloodwork panels? I got a full work up. Hormones, cancer markers, etc. 5+ of my friends went and asked PCPs for it. ALL of them fought back. You don’t need it. Unnecessary stress. Why the hell is this a thing?

@fibrusbroadband do you have a status page or something to track incidents in an area? Would really reduce your phone call volume. When I get a call about a remote relative having no internet it would be in everyone’s interest if there was a status page to check the area

@pcardno @ChShersh @gh_0_5_t Ding ding ding. I bet I could find a problem with a similar level of domain knowledge required that the OP wouldn’t be able to do unless exposed to it before, and I bet he wouldn’t like it if I unhelpfully said his failure to do it means he can’t code.

Here’s a proper 1-hour SWE interview question to filter out people who don’t have basic coding skills: Implement a pure function from scratch to calculate the day of the week by year, month and day.

@DrJimFan First time reading of the “bitter lesson” so might be a bit of a silly question, but does moving focus to compute on inference mean we can get better model performance out of small models w consumer GPUs given that they tend to be still great compute resources but memory poor?

OpenAI Strawberry (o1) is out! We are finally seeing the paradigm of inference-time scaling popularized and deployed in production. As Sutton said in the Bitter Lesson, there're only 2 techniques that scale indefinitely with compute: learning & search. It's time to shift focus to the latter. 1. You don't need a huge model to perform reasoning. Lots of parameters are dedicated to memorizing facts, in order to perform well in benchmarks like trivia QA. It is possible to factor out reasoning from knowledge, i.e. a small "reasoning core" that knows how to call tools like browser and code verifier. Pre-training compute may be decreased. 2. A huge amount of compute is shifted to serving inference instead of pre/post-training. LLMs are text-based simulators. By rolling out many possible strategies and scenarios in the simulator, the model will eventually converge to good solutions. The process is a well-studied problem like AlphaGo's monte carlo tree search (MCTS). 3. OpenAI must have figured out the inference scaling law a long time ago, which academia is just recently discovering. Two papers came out on Arxiv a week apart last month: - Large Language Monkeys: Scaling Inference Compute with Repeated Sampling. Brown et al. finds that DeepSeek-Coder increases from 15.9% with one sample to 56% with 250 samples on SWE-Bench, beating Sonnet-3.5. - Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters. Snell et al. finds that PaLM 2-S beats a 14x larger model on MATH with test-time search. 4. Productionizing o1 is much harder than nailing the academic benchmarks. For reasoning problems in the wild, how to decide when to stop searching? What's the reward function? Success criterion? When to call tools like code interpreter in the loop? How to factor in the compute cost of those CPU processes? Their research post didn't share much. 5. Strawberry easily becomes a data flywheel. If the answer is correct, the entire search trace becomes a mini dataset of training examples, which contain both positive and negative rewards. This in turn improves the reasoning core for future versions of GPT, similar to how AlphaGo’s value network — used to evaluate quality of each board position — improves as MCTS generates more and more refined training data.

@SJAMcBride I’ve stopped speaking first beyond a mumbled hello when answering calls I don’t know. I firmly believe the next generation of scamming will be to get you talking so your voice can be recorded and fine tuned on an AI Model to impersonate.

I fell for a very simple online scam which then became more sophisticated. My editor thought that not enough people had read about this so you should hear about my stupidity...his enthusiasm was such that he even took off the paywall. belfasttelegraph.co.uk/opinion/column…

@jburnsmuse @nntaleb Personally, I don’t try to medicate fever away unless dangerously high. Now and again I make an exception if fever is disrupting my sleep to the point where interfering with the fever response is a trade off worth it to get decent sleep.

@nntaleb Have you ever asked about fever reducing medication? I imagine if the fever gets too high, fever reducers would be appropriate, but maybe not for just minor fevers. I’ve thought about asking my Dr., but not every Dr. reads Antifragile.

Turn out that icing does not help. There was no empirical backing. When I asked the doctor if there was evidence icing worked he responded: "you have the nose the size of Cleveland and you are asking for... studies?" From #Antifragile

@propertypal Thanks. Would much rather "off street" be the only option here and then let the viewer of the listing be the judge as to whether or not it constitutes a driveway.

@ac2u Thanks for your feedback, we will pass this over to our dev team to look at.

@propertypal There's probably a better UX for this filter when setting up alerts. I can't rely on every agent being precise enough to choose correctly here. Want to choose both but not sure if it then excludes listings with only one selected.

@Axtiti @fattyfatman @nntaleb I get that Universa only deal with big funds to protect them from black swans, what I don’t get is why there isn’t an S&P500 index fund that holds a fixed percentage fraction of the fund with Universa where ordinary folks can buy the overall fund

@fattyfatman @nntaleb no...

I am fucking fed up with BULL/BEAR 1st order discussions in finance! Spitznagel: "What matters more than my views on this are how Universa's clients are positioned for it — for both a face-ripping rally & for the worst crash since 1929." zerohedge.com/markets/crude-…

@Crashby89 @ChrisAlvino He didn’t call people without complications assholes. He called people who reply about their lack of complications just because they had none of their own assholes.

@ChrisAlvino More people should absolutely be aware of the risk and the higher than advertised complication rate (10-30% according to the article you posted). But calling someone an asshole because they had no complications gains you literally nothing and will cause people to ignore you.

The reason people don't speak up about LASIK complications is bc you get a bunch of fucking assholes in your mentions constantly telling you it was the greatest fucking decision of their life, making you relive the pain and trauma of your mistake over and over again.

@PhinHarper Wishing you the best in your recovery. Do we know how many people have the holes that never exhibit problems? Is there criteria for medical intervention before problems occur for prevention that balances the risk of the intervention itself?

@TslaftwEli @nntaleb “He's like a math detective, always looking for patterns and trying to make sense of the chaos.” I would suppose he would argue that chasing hidden patterns among chaos will likely only find you noise you’ve mistaken as a pattern. Instead, benefit from volatility (antifragility)

I had to ask grok to explain this tweet like I’m a kid: Oh boy, that's a doozy of a tweet! Let's break it down in a way that even a kid can understand. First off, Nassim Taleb is a smart guy who thinks about the world in terms of probabilities and randomness. He's like a math detective, always looking for patterns and trying to make sense of the chaos. In his tweet, Taleb is talking about how he sees history differently from most historians. He thinks history is like a big, unpredictable game of chance, like rolling dice or flipping coins. This is what he means by "nonanticipating stochastic process." It's like a fancy way of saying that history is full of surprises, and we can't always predict what's going to happen next. Now, the "Itô style" and "Stratonovitch integral" part might sound like something from a science fiction movie, but they're actually mathematical tools that help us understand how things change over time. Taleb prefers the Itô style because it's more forward-looking, while the Stratonovitch integral is more backward-looking. It's like the difference between looking out the front window of a car to see where you're going versus looking in the rearview mirror to see where you've been. So, to sum it up, Taleb is saying that he thinks history is unpredictable and full of surprises, and that historians should focus more on what might happen in the future rather than just looking at the past.

I was trying to explain my problem with historians: I believe history is a nonanticipating stochastic process, integrated Itô style, and they view it as a deterministic one or at best a backward looking Stratonovitch integral. #probability

@b1trusss @iamgingertrash 10 years time: “after our success with GPT360… We’re so proud to introduce…. GPT One”

If GPT2-Chatbot is GPT-5, it's over for openai This is potentially a bubble popping event. For Sam's sake, let's hope it's not If it is, I'm loading up on OTM Nvidia puts

@StickDiagrammer @Grady_Booch Thanks for answering, wanted some food for thought 👍

@ac2u @Grady_Booch Similarly, scientific and mathematical proofs are products of reasoning. Some of these can be automated, but only because the rules for formal reasoning are explicitly transformed into lines of code (i.e., not an LLM) (Sorry if this is here multiple times, app is broken)

@StickDiagrammer @Grady_Booch Never really understood how that argument can’t just be dragged into the philosophical arena though with the counter-point of “how do you know that’s not just what your brain is doing with a larger training set?” 🤷‍♂️

@Grady_Booch Even if you ask an LLM to 'explain its reasoning' for an answer, the answer it gives will just be a statistically probable response given the context. It will have nothing to do with how it came to its original answer, and everything to do with the context of the chat.