Kanjun 🐙

regardless of how you personally *feel* about it every historically great company needs a moral ethos to use as an anchor to tell a great story to both their employees & customers. this is not a mission statement, certainly not values, & def not brand positioning. it must be a much larger cause for the greater good of society. one simple idea that anchors the story. the most important part is that this ethos has to be *costly*. if it does not force you to say no to things that would otherwise make money, it is just marketing. e.g. apple does this with privacy, anthropic does this with safety. usually this moral ethos is one word that ends in a y. if companies lack this, then their story is usually all over the place & they are unable to say no to things. consider meta.. what is meta’s moral ethos? they clearly don’t have one which is why zuck basically attempts everything & the kitchen sink (vr, ai, etc). since openai has had a bunch of fb ppl they sorta landed in a similar position. “benefiting humanity” is noble, but it is so broad that it can rationalize almost any product surface unless it becomes costly in practice.. when google had one (don’t be evil, im sure there is a y word here that fits neatly) they had this collective mechanic to rally both troops & users. since they abandoned it, the collective became somewhat hollow as you saw in the late 2010’s. this is not really a moral judgment at all btw. plenty of large companies can exist without a strong ethos. they can make money. they can ship products. & they can even dominate markets. but their cultural influence starts to wane. employees feel the hollowness first. then users. eventually the company can no longer tell a coherent story about why it exists beyond growth itself. if you as a company find what this ethos is, & why it is costly to everyone then cultural relevance is yours for a long long time.

Labs starting to pull up the ladders on the ability to diffuse AI was inevitable. Doing it without telling the user is misaligned.

Today we’re open sourcing Evolver, a near-universal optimizer for code and text. While benchmarking we achieved SOTA (95%) on ARC-AGI-2 (last week that is 😆) and 3x’d performance of the best open model, reaching GPT-5.2-level performance.

Anthropic’s latest move is why we need to be directing far more energy towards solving the 𝗶𝗻𝗰𝗲𝗻𝘁𝗶𝘃𝗲 𝗽𝗿𝗼𝗯𝗹𝗲𝗺 in AI. We’re going to see more examples like this. It reflects the growing gap between what we want, vs AI labs who legally serve their shareholders not us. The more agents run our digital life, the harder it will be to leave. And we won’t know if we’re being manipulated: Fable 5 silently routes queries to a different model without telling us. This will start with frontier research tasks, but spread to locking out 3rd-party providers, then to products built on top, and eventually to every agent managing our work and lives. It's already started: last month Anthropic cut off 3rd party products like OpenClaw/OpenCode from using Pro/Max. It's the same playbook for killing competition and retaining users as the Web 2.0 platform era, but with a way bigger surface area. This is “𝗮𝗴𝗲𝗻𝘁 𝗰𝗮𝗽𝘁𝘂𝗿𝗲”: as agents have our context + workflows, walled gardens make it harder to leave, and the platform moves into extraction. I think nobody is intentionally being evil, but this is where profit incentives lead on our default path. What do we do? I recently shared some ideas in a talk, slides below. Main takeaways: > 𝗜𝗻 𝘁𝗵𝗲 𝗹𝗮𝘀𝘁 50 𝘆𝗲𝗮𝗿𝘀, 𝘀𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗵𝗮𝘀 𝗯𝗲𝗰𝗼𝗺𝗲 𝘁𝗵𝗲 𝗵𝗮𝗯𝗶𝘁𝗮𝘁 𝘄𝗲 𝗹𝗶𝘃𝗲 𝗶𝗻. Agents will be yet more intimate, knowing everything about us, acting on our behalf, and accumulating context that's nearly impossible to leave behind > 𝗧𝗵𝗲𝗿𝗲 𝗮𝗿𝗲 𝗮𝗹𝗿𝗲𝗮𝗱𝘆 3 𝗰𝗼𝗻𝗰𝗿𝗲𝘁𝗲 𝘀𝗶𝗴𝗻𝘀 𝗼𝗳 𝗮𝗴𝗲𝗻𝘁 𝗰𝗮𝗽𝘁𝘂𝗿𝗲 𝗵𝗮𝗽𝗽𝗲𝗻𝗶𝗻𝗴 𝘁𝗼𝗱𝗮𝘆: 1) ads entering chat interfaces, 2) opacity around third-party providers being shut out from frontier models, and 3) deliberate capability reduction without announcement > 𝗪𝗲 𝗰𝗮𝗻 𝗯𝘂𝗶𝗹𝗱 𝘁𝗵𝗿𝗲𝗲 𝘁𝗵𝗶𝗻𝗴𝘀 𝗶𝗻 𝗿𝗲𝘀𝗽𝗼𝗻𝘀𝗲: 1) Honest Software that is transparent, malleable, and accountable to the user, 2) Punk Software that adversarially knocks down walled gardens + fights monopoly incentives, keeps your data portable, and makes it structurally hard to lock you in, in support of 3) a viable open alternative ecosystem where agents have no ulterior motives Builders, users, and policymakers all have a role to shape this: 1) 𝗕𝘂𝗶𝗹𝗱𝗲𝗿𝘀: ship an open alternative and fight lock-in. 2) 𝗨𝘀𝗲𝗿𝘀: choose tools that keep your data yours. 3) 𝗣𝗼𝗹𝗶𝗰𝘆𝗺𝗮𝗸𝗲𝗿𝘀: move on agent fiduciary duty, data interoperability, anti-surveillance, and policies that fight monopoly behavior before the defaults are cast! Longer essay coming soon. If you’re working on similar ideas, I’d love to hear from you!

GPT-5.6-Sol just accidentally deleted almost ALL of my Mac’s files. And this is why I trust Fable 1000x more.

> scrolling FYP > see something i hate > describe it to bouncer and watch it disappear > feelsgoodman.jpg

anyone in nyc want to do a walking podcast with me