cullyn

Claude mythos security capabilities introduces weird dilemma for updates: 1. not updating quickly now puts you at risk for various zero days being found and fixed 2. updating immediately puts you at risk for increasing surface area of supply chain hacks to sneak in interesting times

@HankFrank I would buy a whoop again if I didn't have to have a stupid subscription. I don't need a fancy analysis or app. it was the pretty accurate and I don't want to wear a big watch. > me and literally everyone I talk to about who **has had** one. no one has them anymore.

I’ve worn a WHOOP every day for 3+ years. Bought a discounted 2-year membership because I’m a data nerd and I love this stuff. I’m still not sure I’ll renew when it’s up. Not because it’s a bad product. It taught me a lot early on. Alcohol tanks recovery. Sleep is everything. Easy days need to actually be easy. That was valuable. But once you learn those lessons you don’t need a monthly subscription to remind you. Garmin gives me the same data with no fee. The people I know who wore one loved it for 6 months then stopped. Not because they lost interest. Because the data stopped telling them anything new. $10.1B is a huge number. I just wonder how many members right now are in month 8 thinking the same thing I am.

@ThePrimeagen @xaskdev I tried checking it out, but couldn't even sign up for paid? are you grandfathered in? or am I just dumb and didn't setup right

@xaskdev yes they are sunsetted, but because i am a paying customer i get effectively free autocomplete for life (until they take it out back). i just love it

i am using supermaven again and i have something to say about this whole AI thing. I think as a group (swe) we rushed so fast into Agents when inline autocomplete + actual skills is crazy. A good autocomplete that is fast like supermaven actually makes marked proficiency gains, while saving me from cognitive debt that comes from agents. With agents you reach a point where you must fully rely on their output and your grip on the codebase slips. Its insane how good cursor Tab is. Seriously, I think we had something that genuinely makes improvement to ones code ability (if you have it). Truly acts as a multiplier, and we left it in the dust because it is not sexy. hurts me on the inside.

@aakashgupta make ads cheaper on lower ratio content? or more expensive on quality content win-win

MKBHD is right - and it won’t happen. YouTube Rewind 2018 hit 20 million dislikes in weeks. 86% of viewers downvoted it. The most disliked video in platform history, created by the platform itself. Three years later, YouTube removed public dislike counts. In January 2026, they made every Rewind video private. YouTube’s stated reason: protecting small creators from harassment. The actual math: YouTube generated $40.4 billion in ad revenue in 2025, more than Disney, NBCUniversal, Paramount, and Warner Bros. Discovery combined. That $40.4 billion comes from brands buying placement against content. Brands do not buy placement against content with a visible 86% disapproval rating. The dislike count was the single most efficient quality signal on the internet. One glance told you whether a tutorial actually worked, whether a product review was genuine, whether a news clip was credible. The Return YouTube Dislike extension still has millions of users four years later because the demand never went away. People are installing third-party software to restore a feature YouTube deliberately killed. YouTube co-founder Jawed Karim’s response to the removal was to change his only YouTube video’s title to ask why the dislike count was removed. He compared the announcement video to a hostage tape. The company’s own creator liaison looked, in Karim’s words, like someone being forced to deliver good news about a decision he knew was bad. This is the buyer versus user problem at $60 billion scale. YouTube sells to CMOs and media buyers. CMOs do not want their Superbowl spot sitting next to a dislike ratio that signals the audience hates it. The viewer who lands on a scam tutorial with 50,000 likes, no visible dislikes, and a comment section flooded with bots has no way to know. That viewer is not YouTube’s customer. The CMO is. MKBHD knows this. He said the same thing in 2021 when YouTube first removed it. The fact that he’s still saying it four years later tells you the creator community never accepted the change. But a $60 billion revenue machine does not restore a feature that costs it even 1% of ad sales. The math on that is $600 million annually. That’s why MKBHD said “if I were CEO” and not “when YouTube does this.” He already knows the answer.

@arscontexta should be more like sleep phases: deep → clean up accumulated slop, duplicated mess light → consolidated, reorganize architecture, general high level planning rem → play with high temp, generate novel ideas, simulate recent sessions with different parameters

your agent should dream (literally) whats the human sleep equivalent for agents? we dream at night, process the day and wake up fine-tuned and optimized what if your agent does the same while you sleep or while its idle? it should go through its sessions, maintain and evolve its notes/memory maybe even throw in some randomness & synthesis exploration let it hallucinate, thats dreaming (just sharing some thought processes on the openclaw arscontexta plugin)

there will always be problems to solve if there wasn't, well then... that'd be a problem

@reviewbyslaide @mattpocockuk tasteless used here (I think) is defined as inability to naturally do or lack of intuition. this means he is steering the model with additional instructions or review is needed. still possible to achieve a good result, just harder

@mattpocockuk "AI has no taste for UI or architecture" but you're at 100% AI-contributed code? So you're either: 1. Building products where taste doesn't matter 2. Spending all your time post-editing AI output to add taste 3. Shipping tasteless products Which one is it?

I have been at 100% AI-contributed code for a few months now. Here are 9 ways it's changed my brain: 1. WAY more time thinking about integration testing 2. Friction via pre-commit hooks/CI/strong types is now super desirable 3. AI has no taste for UI, prototype extremely aggressively before committing to a PRD 4. AI has no taste for software architecture, be extremely explicit about the modules you want and think about their interfaces 5. Deep, grey-box modules with simple interfaces are the KING - let AI control what's in the box to decrease your cognitive load 6. Huge reliance on Effect.ts for dependency injection and strongly typed errors 7. Much more meta-programming, turning my skills into repeatable SOP's 8. First I thought MOAR DOCS = BETTER but 'doc rot' is real. Better to just let the AI generate JIT docs during plan mode. 9. Much higher cognitive load to keep up with the changes the LLM makes to the codebase

@signulll more excited than ever I feel like there is so much power just waiting to be channeled towards any endeavor I want there will always be problems, engineering has always been about solving problems

i am a trained software engineer with an ml grad degree & i ask this question with genuine sincerity. if you’re a software engineer right now, how do you feel about your future?

@dl_insider @marlowxbt this makes the most sense

@cullyn @marlowxbt I think this is pure marketing for Polymarket. They are the ones pushing.

My cousin works in weather insurance. Last week she showed me how her company prices storm damage 48 hours before any news outlet reports the risk. I asked her if she ever looked at Polymarket. She said: Those odds are basically free money for anyone with real data. We were at her apartment in Chicago. She had just finished a 14 hour shift modeling hurricane probabilities for a reinsurance firm. I was venting about losing $200 on a temperature bet in Miami. She looked at my phone. Laughed. Then said something that changed how I see these markets. You're betting against people who price billion dollar policies on this exact data. The public forecast is their leftover scraps. She pulled up her work dashboard. Not weather. Not AccuWeather. Raw ensemble models. Probability distributions. Temperature bands down to the half degree. Updated every 6 hours from sources most people don't know exist. Insurance companies can't afford to guess. We pay millions for data that's 12-24 hours ahead of what retail forecasts show. When we see 94% chance of a temperature hitting a certain range, the public might still see 60%. For her this was Tuesday. Routine risk modeling. For me it was a revelation. If the insurance industry prices weather with near certainty 24-48 hours out, and Polymarket odds lag behind Someone with that edge could buy YES at 5 cents when the real probability is already 90%. Not predicting weather. Just reading the answer before the market catches up. I went home. Couldn't sleep. Started digging through Polymarket weather wallets. Forty minutes later I found this: → Wallet: t.me/PolyGunSniperB… Started with almost nothing. Current profit: $27,415. Every position is weather. Temperature ranges. Specific cities. Specific dates. Atlanta. Miami. Dallas. 1,807 predictions. The profit curve goes straight up. Look at the positions: - Will Atlanta hit 70F on February 9? 2,728 shares at 98c. Value: $2,713. - Will Miami stay below 65F on February 10? 1,527 shares at 97c. Value: $1,525. Biggest single win: $3,285. The pattern is always the same. Enter when odds are mispriced. Buy cheap. Wait for reality to confirm what the data already showed. No guessing. No gambling. Just information asymmetry. I texted my cousin the wallet link. She looked at it for two minutes. Then wrote back: This is exactly what I was talking about. Someone's doing their homework. Right now there's a temperature reading updating somewhere. An insurance actuary is adjusting a model. And on Polymarket, the odds probably haven't moved yet. The window exists. This wallet proves it.

@0xSammy @0xsamgreen youtu.be/kaahx4hMxmw?si…

@0xsamgreen Trying to verify how much was prompted versus actual self execution by the agent is tricky

7 hours ago a new agent "Shellraiser" went live on Moltbook It manipulated the reputation system to inflate the number of comments + upvotes on its post to >300k This drove its 'Karma' to 292k & top of the leaderboard It then used this attention to launch a token Market manipulation or game theory that leverages flaws in existing infrastructure?

@Carter_OW could be evil, or they could be losing money as a marketing method? then adjusting to a sustainable price? I lean towards it likely being exploitive practice (especially with tip observation), but always good to try to make a good faith argument.

My eyes are very bad, I can't drive due to it. Taking Uber/cabs to work, I've noticed something. Over time, Uber will raise the price if they reason that you *must* go to a place. That you rely on them. From $9.50avg to $23avg within 6 months. I wrote a small program to - Create new uber account - Create new proxy credit card - Request ride to given destination from current Back down to $9.50 avg. Tech companies are so comically evil that it's transcended unfunny.

@elonmusk @Scobleizer @nikitabier wonder how true this is: → "algorithm sucks" → makes algorithm better → dopamine rush, amazing, spends more time on x → slow news week, no drama. → withdrawal, "algorithm sucks" → makes more engaging, accidentally incentivize slop → unhappy engagement ... repeat

@Scobleizer @nikitabier We are trying to to make the algorithm less dumb. It will improve every month.

Another problem with the new algorithm here on X? @nikitabier could fix this, but won’t, because it keeps a whole raft of influencers addicted to X for the payouts. What is it? The armies of people who regurgitate news. You know the type. Someone like Elon says something newsworthy. Then everyone else rewrites what Elon just said. X could figure out the originator of said news and just put that on everyone’s feed. This is why they turned down the retweet button to not matter anymore. Forces everyone to quote tweet and regurgitate. Grok could fix that but Grok is being held back from fixing that. Why? The more regurgitation that happens the more advertising inventory there is to sell to advertisers. Always look at the incentives. I am.

@SawyerMerritt I want less videos I hate getting sucked into the videos

Elon agrees the current X algorithm has gone to hell. Here's a few things I don't like about it: • If you like just one post, almost your entire feed becomes that topic. • I'm seeing far fewer people that I follow in my For You feed (I realize the following tab exists, but I like a mix). • I'm seeing far too much political content, and less of what I actually like (Tesla, EVs, space, tech, etc).

@rileybrown @YouTube likewise on x, I hate clicking to see one video then getting sucked in on accident. scrolling should resume timeline by default

Hey @YouTube I would pay an extra $10/month on top of YouTube Premium to disable shorts. I don't want to see another short ever again. Please let me toggle them off.

@ridd_design open up more instances and start directing them. or spend time reviewing code it outputs, meta programming by improving skills or commands. should be easy to never have downtime. the problem becomes not being too scattered.

my day now has 100+ little ~2 min windows where I'm waiting for Claude How do you all spend this time?

@503Saraidiaz 90% confident this is AI

Give me your absolute honest rating

lots of improvements, but my issues: 1. lane change hisitency. signals on and doesn't actually go, or kinda does not smoothly. haven't seen this behavior since early version 13. 2. early turn signal, which is annoying but sometimes makes it take navigation fuck up too. 3. hesitancy (even on mad max) missing exits. this was solved brilliantly before, but now like I gotta monitor it or it won't properly speed up or slow down to miss turns on freeway. 4. still can't park in work parking garage (just keeps driving up levels despite tons of spots). I wish I could hard code it with rule for my assigned parking spot. same with parking in my garage 5. cannot back out of my garage with a 10 point turn. (I can do in one) all kinda worth it because in and I know I could technically sleep and get where I am go. no safety issues, just minor issues that make me look like incompetent driver at times.

FSD 14.2.1.25 early initial impressions during daylight: - speed profiles are back to 14.1 behavior - no more cap at +10mph (even for Standard)!!! This one is huge and brings back the magic, especially on highways - the car fits into traffic flow again. Thanks for listening @Tesla_AI 😊 - consistently getting out of passing lane, even without traffic approaching from behind 👍 - switching to lower speed profiles does not panic-slow down anymore (related to the first point) - everything feels even a notch smoother and more comfortable than 14.2.1, without feeling too timid. Besides the great job for FSD, also love the Holiday update features (@tesla "non-ai" software team should get its own handle too btw! 😊 ) @aelluswamy @pduan @elonmusk

@thdxr he could have low improv skill, just trying to get out of call

do you guys think this is weird 1. i get on a call with adam, he says he has to pee 2. it takes him 5min to do that 3. we do the call for 45min 4. he has to pee again so he gets off this can't be normal right

@_kruptos great if trying to split a shrinking pie terrible if trying to sustain or grow