Martin Lynch

@vawkeicodewebz @garrytan I'm the founder of XLR8•4ward. We make an AI-powered underwriting platform (InfiniteUnderwriter) for the P&C insurance industry, which I spent a few decades in. The industry is powered on profoundly obsolete tech. Every day of my career I've wanted to fix that. Now I am.

@nikitabier @levelsio @X Why not incorporate a 4th pop-up with a poll: "Do you like all these pop ups the EU requires us to present to you?" Might provide some interesting data for regulators...

@levelsio @X There are now 3 additional popups for EU users who signup for X. 30% of my time is now spent on EU compliance.

I got a message on @X from the 🇪🇺 EU today!

@Teslarati Inductive charging embedded in select roadways gets you both.

Would you rather have: Faster Charging ⚡️ 🔌 OR Longer Range 🔋 🚗

@RnaudBertrand Great there's no algo (& especially no humans) deciding what's "true" or has "credibility"-that's a feature, not a bug. In a clearinghouse of ideas, presuming readers can reach their own conclusions is essential (vs "I can figure out truth but you can't, so I'll do it for you")

So I spent some time studying the new Twitter/X algorithm today since the latest version was published about a week ago on Github (#updates--may-15th-2026" target="_blank" rel="nofollow noopener">github.com/xai-org/x-algo…). My goal was to answer why so many people have seemingly seen such a dramatic drop in their posts' reach. The first answer, which is actually somewhat unrelated to the ranking algorithm on Github, is the auto-translate feature, rolled out worldwide on April 7, 2026 (x.com/nikitabier/sta…). Before that date, if you wrote in English about, say, the Trump-Xi Beijing summit, you were competing for attention with maybe 5,000 other English-language accounts writing on geopolitics. After that date, your post is competing for attention with other posts on the same topic IN EVERY LANGUAGE ON EARTH. For some topics that do command global attention like geopolitics, that's a very brutal multiplier: you used to be one of 5,000, you're suddenly one of 50,000 (something of that order): MUCH more difficult to stand out. Secondly, the number of followers you have matters far less than it used to: each post now has to earn its audience reader by reader, on the predicted engagement of the post, and how its topic matches what each reader has recently been engaging with. Here is how the algorithm works, in simple terms: when you, as a reader, open your feed, the algorithm doesn't load "posts from accounts you follow." Instead it runs a 2-stage prediction of what posts you're likely to engage with in that very moment. The first stage is the retrieval stage. The system narrows billions of posts on X/Twitter that day down to roughly 1,500 candidates by matching the semantic content of each post - what it's about - against what you as a reader have recently engaged with. Some candidate posts come from accounts you follow; others are pulled from across the platform by pure topic similarity to your recent interests. You can test this retrieval stage easily: start disproportionally engaging with - say - Brad Pitt videos and you'll bit by bit see your timeline flooded with Brad Pitt content, most of it from accounts you've never followed and never heard of. Then there's the ranking stage. Each of these candidate posts for your feed is fed through a Grok-based model that tries to understand if you'll engage with the post. It looks at 15 engagement metrics: 1) P(favorite) — the reader likes the post 2) P(reply) — the reader replies to it 3) P(repost) — the reader reposts it 4) P(quote) — the reader quote-tweets it 5) P(click) — the reader clicks a link in it 6) P(profile_click) — the reader taps through to your profile 7) P(video_view) — the reader watches the video 8) P(photo_expand) — the reader expands an image 9) P(share) — the reader shares it (DM, off-platform, etc.) 10) P(dwell) — the reader stops scrolling and lingers on the post 11) P(follow_author) — the reader follows you after seeing it 12) P(not_interested) — the reader marks "not interested" 13) P(block_author) — the reader blocks you 14) P(mute_author) — the reader mutes you 15) P(report) — the reader reports the post Fifteen predicted actions, each multiplied by a weight, summed: that sum is the score that determines in which priority a post will be seen among other candidates. Please note that posting something with a video or an image can give your post an advantage as 2 actions are specifically for these: video_view and photo_expand. No video or photo and you don't get a score for these. Also, naturally, having a video maximizes the chance that a user will "dwell" on your post to watch it. Also note that 4 of these actions carry negative weights (not_interested, block_author, mute_author and report): meaning that if the model expects a post to generate a lot of negativity, it'll get de-boosted quite dramatically. But note, first and foremost, what's NOT in there: none of the things that, naively, one might think a serious information platform would weigh. There is no P(this post is true and well-sourced). No P(the author actually knows what they're talking about). No P(this person has spent a decade building a body of work that has held up). No P(this account has earned the right to be taken seriously on this topic). No P(the author has a large following from credible people). The model does not seem to care - at all - about any of that. Every post starts from zero. You could have ten years of rigorous, well-sourced analysis behind you - or you could be just an uneducated rando who registered yesterday. To this algorithm, you're both just a bag of engagement probabilities. Now, sure, to be fair, there is a "brand" effect that's not covered by the algorithm: someone who has in fact built a brand will naturally have better engagement metrics because people recognize their account. But that's an indirect, second-order effect. And crucially, it's legacy: those "brands" were built under earlier versions of the algorithm that gave followers and reputation more weight. Lastly, several other features of the new algorithm compound the dilution, none of them visible from outside but all consequential. The May 15 update added an "impression bloom filter," tightening the rule that once a reader has been served a post, the system won't serve it to them again. Before, a strong post could marinate in someone's feed across multiple refreshes and accumulate engagement on the second or third pass. Now it basically gets one shot. Also, your own posts compete with each other. An "Author Diversity Scorer" inside the ranking stage attenuates the score of every subsequent post of yours that ends up in a reader's candidate pool. In plain terms: if multiple of your posts land in a reader's candidate pool, the system shows one at full strength and dampens the others. So don't post several times consecutively on the same topic. And, last but not least, another huge impact on reach is that, in the old algorithm, when someone reposted or quote-tweeted you, your post was broadcast to their followers' timelines - a repost from an account with 100,000 followers was a huge boost. In the new algorithm, that mechanism is vastly demoted: reposts - like every post - need to go through the retrieval and ranking stage mentioned above, so a repost from a big account is a long way from the boost it used to be. This is especially brutal for low-effort quote tweets, which used to function as cheap amplification: now they often can't even clear the retrieval stage - they simply don't contain enough novel semantic content for the system to match them to anyone's interests. So, putting it all together, the reach collapse comes from many forces stacking at once: - Auto-translate makes your posts compete for attention against an order of magnitude more content - The retrieval stage matches posts by topic, not by who follows you - The ranking stage scores purely on predicted engagement with no weight for credibility, expertise, or track record - The bloom filter narrows every post's window to one strong shot - The diversity scorer penalizes prolific posting - Reposts no longer carry much distribution power Each of these alone would dent your reach. Combined, they amount to a complete reset: your audience that you built painstakingly over years basically doesn't matter much anymore, and it's much - much - harder to stand out even if you're a big account. People structurally rewarded by this algorithm are folks who: - Post visually (videos/images) - Post on globally popular topics because they clear the retrieval stage easily - Provoke strong emotional reactions - likes, replies, reposts - Don't care about accuracy or seriousness because the algorithm doesn't measure it - Don't care about their existing audience because every post is judged in isolation anyway In short this new algorithm, like so many on social media, is all about maximizing whether people will engage with something - not about whether they should.

Strange qualities I've encountered all too frequently among doctors: - incuriousity: a genuine lack of desire to understand - odd to find in a profession where curiosity is a prerequisite to excellence. - gatekeeper: thinking their job is primarily to deny my requests. I had to argue at some length with a doctor to get an Apo B test. After twice refusing he finally relented, but then said, "I probably won't even look at the result." Nothing like indifference to convince me of your professionalism...

I went to the doctor asking for blood tests, as I’ve been experiencing some minor MCAS and histamine issues. She denied certain tests because she “didn’t think they were necessary,” prescribed me Prednisone, and sent me on my way. I said I didn’t want to be taking steroids and she ignored me, told me to just take it, and explained none of the side effects, which can potentially be very severe and long lasting. Absolutely insane. This was a great reminder of how important it is to take your health into your own hands and do your own research.

@TeslaTopics Exactly the same for me on 14.2.2 (I think) - identical circumstances: the truck lost its nerve halfway through a left turn with oncoming traffic, and stopped half in the path of a vehicle. I powered through the turn. It had been over a year since I had a safety intervention.

This is the first time I’ve ever had to take over with a critical intervention in a long time. I have never experienced this kind of hesitation on a left turn before, even in earlier versions, where it would stop mid-turn and block the path of a car going straight. I hesitated to take over, thinking there was maybe someone crossing, but there wasn’t. What happened here @elonmusk @aelluswamy @Tesla_AI?

@mikepat711 That's a pretty good heuristic, because it requires you to have a deep belief in the company's mission, and it's unlikely you get that from companies likely to produce mediocre results. I operate the same way.

I don't consider investments based on making money. I consider investments based on which companies I'm thrilled to go broke with. So far this has made me a lot of money.

@tobi It's another data point that makes you wonder the degree to which many jobs are essentially a UBI scheme in disguise, poor performance being no barrier to continued employment. It seems to concentrate disproportionately (though certainly not exclusively) in government.

The idiot tax

@garrytan Exactly right & it's where we're at: working extremely closely with early customers who are surfacing issues, the manual correction of which daily improves our AI-powered P&C underwriting platform. It's the best & fastest possible feedback loop (tho occasionally very painful).

Everyone thinks "do things that don't scale" is about building relationships with early users. Yes AND it's about generating mistakes at maximum density. When you're doing everything manually (onboarding, support, delivery) you hit errors every hour. Each error teaches you something the dashboard never will. The manual work IS the learning. Automate too early and you freeze your ignorance in code (and now markdown).

@donalddhoffman On point. Conscious beings act as XOR gates turning an otherwise undifferentiated universe into all the things we experience. Prior to encountering a conscious agent the notion that two things might be different from each other (or that there are two things) does not exist.

“philosopher Victoria Trumbull argues that consciousness provides the space for the possible to become real. In a universe driven by laws, consciousness is the creative factor” iai.tv/articles/consc…

@DrJStrategy Excellent post, but the better question is whether central banks the world over ought to fix the price of money at all. In no other realm is it considered smart to fix prices, for the obvious reason it removes price discovery signal to suppliers & consumers, distorting markets.

Let’s be honest. Warsh at the Fed. Kevin Warsh’s arrival at the Federal Reserve is not a personnel change. It is a regime change attempt inside an institution built to prevent one. A supply-sider now runs a central bank hard-wired for Keynesian demand management, and the machine is already resisting the new code. The next mistake is visible in plain sight. Keynesians on Wall Street and inside the Fed are treating a supply shock as if it were a demand boom and calling for tighter money. This is dogma masquerading as seriousness. A chokepoint in the Strait of Hormuz, a jump in energy prices, and a cost shock rolling through transport, food, and manufacturing are not evidence of overheated demand. They are evidence of a damaged supply side. Monetary policy cannot reopen a shipping lane. It cannot pump more oil. It cannot repeal geopolitics. It can only crush demand somewhere else, usually with a lag, and usually in the most interest-rate-sensitive corners of the economy first, housing, commercial real estate, capital spending, and durables. Those sectors did not close the Strait. They are simply first in line to pay for the Fed’s intellectual mistakes. That is the Keynesian reflex in its purest form. Every price spike becomes “inflation.” Every inflation scare requires a rate move. Every rate move is advertised as proof of resolve. It is nonsense. A change in relative prices caused by a supply shock is not the same thing as an inflationary spiral. Pretending otherwise is how central banks turn an external shock into a domestic recession. Machiavelli explained why change is so hard. The innovator makes enemies of everyone who did well under the old order and wins only lukewarm defenders among those who might benefit from the new. Christensen gave the same warning in corporate language. Incumbent institutions kill disruptive change because their processes, incentives, and prestige are built around the existing model. That is the real problem Warsh faces. The resistance is not incidental. It is structural. The test for Warsh is not whether he can sound tough on television. It is whether he can resist the Wall Street catechism that every supply shock must be met with tighter money. If he hikes rates into a supply-driven price spike to prove his anti-inflation credentials, he will not have broken with the Keynesian regime. He will have submitted to it. This is not the 1970s. Expectations are not unanchored, and the productive economy is already scarred by years of policy excess, fiscal decadence, and institutional bias. The hope is that Warsh understands the difference between inflation and a supply shock, ignores the Keynesian pundits, and refuses to compound one policy error with another.

@tobi Tobi on point as usual. All taxes are paid by people - there are no taxes that corporations pay. All corporate taxes are ultimately borne by the customer. Corporate taxes are a very effective way to tax people more while convincing them they aren't actually paying.

Our regulators have lost the plot. Literally, figuratively, and metaphorically.

@D0miniqZ @twentythrill Thanks, that's exactly what we're exploring: locally hosted open models. If performance is even close to last gen models then the freedom & control are worth the upfront cost for us. Google's recent moves have really clarified our thinking on this (likewise the 429 responses).

@MartinMLynch @twentythrill Ngl if u switch to Claude, you’ll feel the same, if you switch to GPT maybe a bit more usage as they’re more token conscious, but still you’d pay fortune to keep coding. The only good recommendation is local- buy once, then self host and you got no quota limits.

Gemini 3.1 Pro is completely broken since the platform update. We’re paying more for tighter compute limits, only for the model to hallucinate wildly and completely fail to read basic chat context. I don't know what they did to the backend, but they massacred my boy.

@_mohansolo It took 1 hour of coding with Gemini to hit the rate limit I didn't previously hit in 18 hour coding marathons. Between that & general regressions to Gemini 3 series (which struggles to follow instructions & produces a lot of breaking code) I'm exploring alternatives.

An update: we’re 3xing the rate limits for Gemini models across all paid tiers in Antigravity and resetting everyone’s Gemini quota for the week. We understand some people hit their rate limits quickly and wanted to respond fast. Lots more to come and enjoy building!

@GigaBasedDad Avatar completely lost me when Sigourney Weaver's character (a scientist) started smoking a cigarette in the limited oxygen environment of a spaceship. I can only suspend disbelief so far...

Why did the movie Avatar make so much money? I still don't get it... the movie was incredibly mid...

@tobi We are making strong headway in our small corner of the AI world (AI-powered P&C insurance underwriting)! Even very large insurers have been willing to engage & been eager to move forward rapidly. It's been a welcomed surprise! But your point stands: we all need to accelerate.

We need Canada to go all in on AI adoption

@JasonBotterill Gemini generally, but the 3 series in particular, refuses to follow instructions. It doesn't matter if you include them in a markdown file, system instructions, prompts, etc. It does its own thing. It's maddening. Threats & profanity, used liberally, help a bit, but not much.

Not even kidding 5 minutes with Gemini 3.5 flash was probably more dangerous than unprotected sex. I think Google didn’t actually test this model with real people as it does things you didn’t ask for constantly

Gemini 3 series has been quite a frustrating disappointment for coding: it routinely generates massive amounts of errors, changes variable names arbitrarily & has a penchant for breaking changes. 2.5 was peak Gemini thus far (though none seem to match other models for coding).

@wholemars Once Tesla releases "unsupervised" the liability for accidents (& maybe traffic violations) presumably moves from the driver to the company. My guess is we'll have "unsupervised" but labeled "supervised" for some time while the financial impact of the change at scale is modeled

Tesla is quietly rolling out FSD Unsupervised to customer cars without anyone noticing That’s what it feels like as I sit here tweeting while my car drives me over a mountain

Article: a16z.com/need-series-c-…

Excellent insights from Andreesen Horowitz (a16z) to help us think about how AI adoption might play out (link in reply). We tell all our customers our platform moves their bottleneck from the bottom of the funnel to the top: it becomes about how much opportunity you can feed it, since processing those (complex) opportunities becomes trivially easy. We too think the value of lead generation is going to skyrocket.