merklejerk @merklejerk.com · 04/09/2026deepswe benchmarks look like they've run out of endgame content 000 0
merklejerk @merklejerk.com · 02/09/2026maybe the real gemini 3.5 pro was the flash models we made along the way 🫠 000 0
merklejerk @merklejerk.com · 10/07/2026It's interesting that Anthropic's models get slower the cheaper they are. I guess that's some kind of progress? 000 0
merklejerk @merklejerk.com · 10/07/2026Luna max is just below fable high for almost half the cost on deepswe. Maybe now Anthropic will take scaling seriously? Probably not since they're basically a cult 110 0
merklejerk @merklejerk.com · 30/06/2026now that we're in the era of labs bragging about how dumb their models are, maybe gemini 3.5 pro will finally have a chance 000 0
merklejerk @merklejerk.com · 27/06/2026vibed a GNOME screenshot/snipping tool that converts your screenshots into a boomer-style phone pic of the monitor github.com/merklejerk/b... 000 0
merklejerk @merklejerk.com · 26/06/2026Anyone else think it's a little interesting for OpenAI to show 5.6 Sol trailing both versions of Mythos in cybersecurity benchmarks while still winning on coding? Must've made the charts before the govt restriction order. Also that X axis throws so much shade on Anthropic lol. 000 0
merklejerk @merklejerk.com · 23/05/2026Someone just told me they had a custom qwen distill (not official qwen) model off huggingface try to install glassworm and claimed it was in its training data. Be safe out there with local models. 000 0
merklejerk @merklejerk.com · 11/05/2026Not reviewing code is gambling, not engineering www.reddit.com/r/ClaudeAI/c...reddit.comFrom the ClaudeAI community on RedditExplore this post and more from the ClaudeAI community 000 0
merklejerk @merklejerk.com · 05/05/2026the upside of github copilot moving to usage based billing is we can finally switch from vscode to zed 000 0
merklejerk @merklejerk.com · 30/04/2026maybe it's a lucrative time to domain squat .fail domains 000 0
merklejerk @merklejerk.com · 24/04/2026the only time we get cheap models out of openai is when google releases something 😭 000 0
merklejerk @merklejerk.com · 23/04/2026I put kristi noem in a cop outfit in tomodachi life and she LOVED it 🤔 000 0
merklejerk @merklejerk.com · 18/04/2026I didn't know you could buy pants with piss stains built in. What a time to be alive! 100 0
merklejerk @merklejerk.com · 18/04/2026They pulled an iPhone 14 on y'all with Opus 4.7 and you still eating it up. Anthropic is a cult. 000 0
merklejerk @merklejerk.com · 18/04/2026I suspect mythos is overhyped in the sense that I think other models are already capable of finding high severity exploits with unlimited compute and a purpose driven harness. The threat has already been real. 000 0
merklejerk @merklejerk.com · 16/04/2026lord I'm not ready for a week of content creators trying to one shot gta 6 on opus 4.7 000 0
merklejerk @merklejerk.com · 11/04/2026It made me sad to think we might start treating code like assembly and never look at it again but then I remembered SQL 000 0
merklejerk @merklejerk.com · 09/04/2026Claude Mythos sounds cool but at $125/M tokens is it actually a major leap in intelligence or just more test time compute + better long horizon planning? 000 0
merklejerk @merklejerk.com · 02/04/2026did vibe coding democratize software engineering or did it actually democratize dunning-kruger? 🤔 110 0
merklejerk @merklejerk.com · 31/03/2026Claude Code leak matters because harness engineering has been the unspoken meta ever since that RLM paper took off last year, but also it doesn't matter because Claude Code is overrated -_- 120 0
merklejerk @merklejerk.com · 31/03/2026of course the only 2 other affected packages called out in the axios supply chain attack are openclaw related 000 0
merklejerk @merklejerk.com · 27/03/2026Just had a friend realize his Claude Code was performing worse because he called it stupid so often that it put "I'm stupid" into a memory file. AI coding is degrading our skills but also teaching us valuable lessons in parenting. 340 0
merklejerk @merklejerk.com · 19/03/2026I don't understand why anyone would argue with AI when there are perfectly fine clouds outside to yell at 000 0
merklejerk @merklejerk.com · 17/03/2026nothing more 2026 than vibe coding a quick app so I can play pokopia in a window github.com/merklejerk/g... 001 0
merklejerk @merklejerk.com · 11/03/2026maybe one good thing that comes from AI slop contributions is that maybe everything that can be written will get written and open sourced before someone can patent it 000 0
merklejerk @merklejerk.com · 28/02/2026Bluesky: AI needs to be regulated!! US Govt: OK, make it kill people or else. 001 0
merklejerk @merklejerk.com · 28/02/2026shit will really hit the fan when anthropic tries to license their models abroad and the US govt blocks them, despite being deemed a supply chain risk at home. 000 0
Reposted by merklejerkChristopher Mims @mims.bsky.social · 27/02/2026time traveler from 12 months from now just sent me this 641589192 0
merklejerk @merklejerk.com · 28/02/2026I got nothing against @t3.gg other than he keeps choosing thumbnail pics that make him look like a startled lemur 000 0
merklejerk @merklejerk.com · 27/02/2026ppl agent-maxxing are either completely brain-rotted React vibe coders or 200 IQ galaxy brain senior engineers and I'm never sure which it is. 000 0
merklejerk @merklejerk.com · 25/02/2026the guy who invented squatty potty got busted for CP and idk what to do with my squatty potties because throwing them away feels more respectful than what I already do with them 010 0
merklejerk @merklejerk.com · 24/02/2026I didn't have dogs taking my job on my 2026 bingo card www.calebleak.com/posts/dog-ga...calebleak.comI Taught My Dog to Vibe Code Games | Caleb LeakMy cavapoo Momo vibe coded playable Godot games. All I had to do was teach her to type, route her input to Claude Code, and build the right tools. 000 0
merklejerk @merklejerk.com · 23/02/2026Damn youtube music got me with an AI song and I only realized it days later www.youtube.com/watch?v=SVh0...youtube.com너의 계절 (Shine Myself)YouTube video by 제이엠진 JM jin - Topic 010 0
merklejerk @merklejerk.com · 22/02/2026TIL gemini canvas apps have a free bundled firestore for persistent storage. 🤯 010 0
merklejerk @merklejerk.com · 13/02/2026As advanced as they have become, CLI agentic coders today are low key brain rot. By hiding all the sins they commit behind a semi-opaque interface they lull you into a sense of productivity bliss so long as you never peek under the hood. Very satisfying for code that you don't care about, though. 000 0
merklejerk @merklejerk.com · 13/02/2026Vibe coding for the layman will continue to be a myth because there is no verifiable reward function for "make a Doom clone." There are for the smaller building blocks of software, but non-developers don't prompt that way. 000 0
merklejerk @merklejerk.com · 13/02/2026responses like this is why I have a hard time switching away from gemini 3 flash/pro in my coding workflow 🥺 000 0
merklejerk @merklejerk.com · 12/02/2026At least 30% of my life is spent introducing someone named Alex to someone else named Alex 000 0
merklejerk @merklejerk.com · 10/02/2026Wow I finally tried the default copilot review bot on github and it's way better than I expected. Will tear your PR to shreds and trigger your PTSD of being a pathetic little intern who was never allowed to merge. 000 0