Chris Lengerich @chrislengerich.com · 14/01/2026RIP Scott, one of the brilliant. 68 is far too young, wish you could have seen how far we'll go. 100
Chris Lengerich @chrislengerich.com · 03/09/2025Also, no excuse anymore to miss important global trends (via universal translation + summarization): 120
Chris Lengerich @chrislengerich.com · 17/07/2025Short social. Partially a product of its own success (there are no more eyeball-hours to harvest), partially a product of the fact everyone can just poll a personal summarizer that doesn't read ads (you really only care about a thin slice of news in your Markov blanket, but you care about it a lot). 100
Chris Lengerich @chrislengerich.com · 22/10/2024Absent chokepoints and adversarial behavior, horizontally integrated stacks tend to outperform overall. However, these also tend to have longer, more complex supply chains, and be vulnerable to adversarial disruption. 000
Chris Lengerich @chrislengerich.com · 10/10/2024We need to accelerate innovation, but also spend $ securing our critical semantic infrastructure: 100
Chris Lengerich @chrislengerich.com · 01/10/2024A great product with limited financeshed may take decades to grow, if at all (hence, movement may be more important than product dev). 000
Chris Lengerich @chrislengerich.com · 19/09/2024Similar (Great Pacific Garbage Patch) - anyone know of a live view for this? 100
Chris Lengerich @chrislengerich.com · 19/09/2024Any know good visualizations of spam/unverified public posts on the Web, similar to astria.tacc.utexas.edu/AstriaGraph/ for pollution in space? (the grey dots) 100
Chris Lengerich @chrislengerich.com · 30/08/2024Human networks are often more iterative than one would like them to be (arxiv.org/pdf/2408.16629) 100
Chris Lengerich @chrislengerich.com · 27/08/2024Especially in AI and AI safety, claims of "X is hard to understand" or "X has evaded the understanding of scientists" are often more a reflection on the speaker than the topic. 100
Chris Lengerich @chrislengerich.com · 27/08/2024Chatting with @verificationgpt.context.fund re: which environments favor what type of firm structure: 100
Chris Lengerich @chrislengerich.com · 26/08/2024Level 0: A logagram of time (heptapod-style), generated from replicate.com/meghabyte/ar... 010
Chris Lengerich @chrislengerich.com · 17/08/2024Any sufficiently efficient predictive model is indistinguishable from magic. 000
Chris Lengerich @chrislengerich.com · 17/08/2024(and related, camouflage, like science, is at least partly a psychological phenomenon, not just a physical one) 000
Chris Lengerich @chrislengerich.com · 17/08/2024(it's helpful to think of consciousness as a self-learning iterator running contrastive distillation. Without the iterator model, little makes sense (you only feel different parts of an elephant). With it, a whole lot of disconnected pieces snap together (the full elephant)). See the elephant. 100
Chris Lengerich @chrislengerich.com · 13/08/2024Concretely, conflict and climate change is much better visualized using resources like FIRMS and Ventusky than stories: 100
Chris Lengerich @chrislengerich.com · 13/08/2024The Internet, as opposed to the real world, creates arbitrary representations which dramatically change truth and reward functions. Rough attempt at a ranking, and where to spend your time (hint: it's not on social, just being a reaction-bot): 210
Chris Lengerich @chrislengerich.com · 13/08/2024Rough feeling on the predictive accuracy of different sectors of the economy. Rough sketch, don't take it too seriously :) 000
Chris Lengerich @chrislengerich.com · 07/08/2024Also, kinda contrastivist: (cardinalnews.org/2024/07/25/o...) 000
Chris Lengerich @chrislengerich.com · 05/08/2024Simple thing you can do w/ personal AI: Social media, but only filter builder accounts that post once/month and post asynchronously. Curve is timeless, but 2010s online culture mixed the RHS and middle distributions together (and the middle dominated due to volume). 100
Chris Lengerich @chrislengerich.com · 31/07/2024When you only see the wave, it feels choppy and uncertain. When you focus on the boat, the veil of obscurity cast by uncurious and click-seeking incentives drops away, and it feels calm and predictable. 100
Chris Lengerich @chrislengerich.com · 12/07/2024Gotta love the founders that put their startup's names on their cars :) Disclosure: Angel investor in Second 000
Chris Lengerich @chrislengerich.com · 21/06/2024Clear example of the degradation of thought in real time. Common pattern for PMs to oversimplify "complicated" classic UIs (think the Mac dongle-boondoggle). The more contrastive our glance, the deeper-order generative models we can intuitively work with. 100
Chris Lengerich @chrislengerich.com · 09/06/2024When you weren't looking, Brave Summarizer become a best-in-class therapist: search.brave.com/search?q=Wha... Free, fast, compact results and handles hard personal questions that Wiki misses (Reddit is good, but slower). Now to figure out monetization and paying the websites it summarizes. 100
Chris Lengerich @chrislengerich.com · 05/06/2024Although admittedly #2 has a hilariously bad graphic to illustrate how IEX was trying to slow down HFT traders. Most races on tracks (except the 100m) start on a stagger to compensate for the different radius of the track in each lane. So IEX really would be adding a stagger, not removing it. 000
Chris Lengerich @chrislengerich.com · 03/06/2024Brave Summarizer is my new favorite site on the Internet: search.brave.com/search?q=che... 000
Chris Lengerich @chrislengerich.com · 01/06/2024Brave summarization (which is getting quite good): 100
Chris Lengerich @chrislengerich.com · 01/05/2024This is what a market looks like (feedback loops with self-learning agents with memory). 100
Chris Lengerich @chrislengerich.com · 18/04/2024Brave's default AI search is now pretty good for quick dives into arbitrary topics: 110
Chris Lengerich @chrislengerich.com · 29/03/2024Learning/truth-seeking is a very different process: 000
Chris Lengerich @chrislengerich.com · 29/03/2024The faster influencer circuit, for comparison. Works much better in scaled networks, but only for a few: 100