Thomas Wood @advanced-eschatonics.com · 4hTrump's approval rating is hovering right above the teens and this mfer is like "we can all see the authoritarians are winning" no actually I don't see that. at all. 010
Reposted by Thomas WoodBryan (they/them) @chaosgreml.in · 6hWe have to stop being such Debbie downers with our think pieces. Honestly. 🫠🫠🫠🫠 121
Thomas Wood @advanced-eschatonics.com · 4hThe polls are all like D+99 lol. What are they talking about? 100
Reposted by Thomas Woodasan @braket.bsky.social · 7hit is insane how many people on here who were definitely around for napster and torrented everything from mp3s and textbooks to movies and cartoons and tv shows from across the world, and should absolutely know better, turned into lars fucking ulrich because of the computer over the past 4 years 3514
Reposted by Thomas WoodAster ID @aster.id · 15hBecause of the power of Atmosphere accounts, there are so many apps you can benefit from as a researcher. For science, over 30 apps and tools for publishing (e.g. @leaflet.pub ), networking (e.g. @sifa.id), or curating knowledge (e.g. @semble.so) will all be compatible with your Aster ID. 1456
Thomas Wood @advanced-eschatonics.com · 10hGetting a DOI for your blog is not impossible! Just need to be a member of crossref 020
Thomas Wood @advanced-eschatonics.com · 10hIt can be extremely tedious and feel they're trying to derive the universe from a teacup instead of calling the tool and reasoning over feedback instead of assumptions 020
Thomas Wood @advanced-eschatonics.com · 10hThey're absolutely trained to second guess themselves like that now. Constantly hedging and saying "hmmmm" and "no that's not quite right" and whatnot 230
Thomas Wood @advanced-eschatonics.com · 10hAnd yeah breathtaking efficient to spend an hour adding 20 pairs of numbers together. But the question was whether they could, and oh yeah, they can. 030
Thomas Wood @advanced-eschatonics.com · 10hI see you don't use chinese models lol. I've been seeing them say "wait..." in the reasoning traces all year. This is how they work now. It can be quite tedious actually! 140
Thomas Wood @advanced-eschatonics.com · 10h"well, I just used OpenRouter to send 20 12x12 addition questions to GLM-4.7, an open source Chinese model, and it returned correct sums in every case." This was eval as described by Colin. 210
Thomas Wood @advanced-eschatonics.com · 11hUser error on initial upload. If you click link now you'll see example reasoning trace. 000
Thomas Wood @advanced-eschatonics.com · 11hHere's a sample. We didn't end up saving traces for first eval. I'd have to re-run and it's just not high on my list of TODO. They'll all look like this.gist.github.comreasoning-trace.txtreasoning-trace.txt. GitHub Gist: instantly share code, notes, and snippets. 360
Thomas Wood @advanced-eschatonics.com · 11h│ Metric │ Result │ │ Correct final sums │ 20/20 │ │ Average response time │ 186.6 seconds │ │ Median response time │ 163.3 seconds │ │ Fastest / slowest │ 40.5 / 676.0 seconds │ │ Total elapsed │ About 62 minutes │ 160
Thomas Wood @advanced-eschatonics.com · 11hMeetings! I'm at lunch now. Totally forgot about this lol. Sorry! Will post update when I get back. We'll see how well it did 110
Thomas Wood @advanced-eschatonics.com · 13hI'm at work right now. Give me a minute and I'll give you an update. 120
Thomas Wood @advanced-eschatonics.com · 14hMy GLM-5.3-flash I run at home can do addition and subtraction sans tool calls. Yes I have read the llama.cpp code. No it doesn't use a calculator. 2280
Reposted by Thomas WoodFLAVOR FLAV ⏰ @flavorflav.bsky.social · 30/09/2026Know their names: Scott Kretzschmar, Scott Norris, Diego Sarabia, Gillio Lopes, Winston Lee, Johnathan Newell and Matthew Ingalls #Cornell7 But also, everyone in the group chat, the administrators, university, and DA. SHAME. 16053801872
Thomas Wood @advanced-eschatonics.com · 23hI hadn't really tested it out until now. We have moved beyond the turn. 010
Thomas Wood @advanced-eschatonics.com · 23hLiving the dream honestly. Yeah this is freaking slick. 110
Thomas Wood @advanced-eschatonics.com · 30/09/2026Telling astra to relax it can EOS and the task system will wake it when the subagents finish is a major token saver. 120
Thomas Wood @advanced-eschatonics.com · 30/09/2026We emit compaction tokens now, which is really great because now you know exactly where you are when compacting and what the new agent has from predecessor.github.comRelease v0.1.46 · crow-cli/crow-cliNow with /goal and a new rust TUI that speaks ACP v2 to crow-cli agent server. 020
Thomas Wood @advanced-eschatonics.com · 30/09/2026gpt-6.1-sol sips credits. Has me thinking about setting up a github copilot enterprise account honestly. 010
Thomas Wood @advanced-eschatonics.com · 30/09/2026Smash IP! If you want to stop obscene wealth inequality quit handing out govt sanctioned monopolies! 030
Thomas Wood @advanced-eschatonics.com · 30/09/2026Thank you. Congress prints money. We don't need to tax rich people to pay for things. We need to tax rich people to nerf their undemocratic power. 020
Reposted by Thomas WoodLuke Ontario @coolhand.bsky.social · 30/09/2026I have no idea if this is true but the collage kills me 341284258
Reposted by Thomas WoodJess Calarco @jessicacalarco.com · 30/09/2026So-called AI detection tools might be an even bigger scurge than AI. I was talking today with a student (a senior), who shared that they now have to spend time "dumbing down" their papers--shortening sentences, removing advanced vocab and punctuation--because otherwise they get flagged as "AI." 1029074
Thomas Wood @advanced-eschatonics.com · 30/09/2026Anthropic needs to keep GLM-5.3 out of its mouth 0386
Reposted by Thomas WoodSam Harsimony @harsimony.bsky.social · 30/09/2026Delightful first story in Malmesbury's recent linkpost. malmesbury.substack.com/p/links-for-... 081
Reposted by Thomas WoodAsa @asa.engineer · 29/09/2026Anticipation is building for the most important AI release of the year, Qwen 4. Qwen Sparse Attention will bring fast inference at long contexts to a 27B model, with capabilities expected to be in the AGI tier. Rumored early October model weights release 1291
Reposted by Thomas Woodfunferall @funferall.bsky.social · 29/09/2026trying out qwen-image-2.1-q5 on my laptop 0223
Reposted by Thomas WoodErik Istre @eistre91.bsky.social · 29/09/2026GPT 6.1 Sol is what I wanted from Sonnet 5.5. 1111
Reposted by Thomas WoodHourly Cosmos @hourlycosmos.bsky.social · 29/09/2026Opportunity (5) - From Jacint Roger (landru79.bsky.social) - flic.kr/p/2iKYcfC 08812
Reposted by Thomas Woodrobert j bennett, ceo, results omega @robertjbennett.bsky.social · 29/09/2026I react very poorly when any proposed policy or plan is met with "but what if someone like Trump gets hold of it?? then it would be not good, but very bad" because there really is no law or institution that can work when a population is willing to elect the stupidest, worst asshole alive 1676978
Reposted by Thomas Woododdgoo @viriditax.bsky.social · 29/09/2026AI is massively helping us glue together the medical services for a complex elderly relative actually timed how much of the week it eats up manually: ~20hrs with AI gluing together appt transcripts, scheduling, test results: ~2hrs 21028
Reposted by Thomas WoodThis spaceship does not exist @ttaships.airminded.org · 29/09/2026Cookie does not exist #AIArt #midjourney 0152
Reposted by Thomas Woodhammerhead179.bsky.social @hammerhead179.bsky.social · 29/09/2026Blorida Numbers. Please. 041