@omarsar0@Recursive_SI the 262k context on something w only about 2.5b active is kinda wild. fwiw the ollama build is macos only for now so on a mac its basically ollama run flint and ur set, steps here https://t.co/ldG8Noct8J
tesla says optimus production started on its high volume lines in fremont, and calls them the first production quality bots its built https://t.co/YcodvSxz9v
anthropic shipped a new api version and new major sdks yesterday. on it temperature, top_p and top_k now come back as an error so look at your calls before you bump the sdk https://t.co/40wZVxcdo9
meta booked a $9.89 billion charge for the teen safety settlement with the states, right about the $10 billion it said in august. now it says payments could come to as much as $18.59 billion https://t.co/EOmYRQVgbJ
@TheHumanoidHub sim helps but the cheap trick everyone is chasing is human video. figure taught helix to move around homes with only first person video of people, no robot demos, and tesla trains optimus on human video too https://t.co/jCU8AvhEZj
@TheHumanoidHub worth holding next to the 10q when it drops tho. last quarter the call was never been more optimistic and the filing showed operating income down 57% to 398M.
calls and filings dont always match https://t.co/SYzdEJJbgu
@TheTranscript_ worth holding next to the 10q when it drops tho.
last quarter the call was never been more optimistic and the filing showed operating income down 57% to 398M.
calls and filings dont always match https://t.co/SYzdEJJbgu
@QuiverQuant not real time at all. pelosi household intel calls were bought may 29 and filed jun 23, and the bloom energy buys from jul 24 were filed aug 21. u always find out weeks later https://t.co/XBAUWMjA6K
@K0iaK0ia@omarsar0 the hard stops part is the right call. same reason i like setup tools that show everything first and ask before changing anything, agents should work the same way. bunch of recipes built like that here https://t.co/4IVPqHb0mV
@yuan_minh5988@omarsar0 same ngl, what kinda helps is not loading the whole memory file at all, just embed each note and pull the 3 or 4 closest to the task so stale stuff stops crowding in. tiny free example https://t.co/2AePSyBW2j
@MetariLabs@TheHumanoidHub agree the source matters a lot. figure taught helix to move around homes using only first person human video, filmed in brookfield homes as people went about their day, no robot demos https://t.co/F9swgY1JaZ
@arimedai@IlirAliu_ google already splits it kinda like that, gemini robotics ER 2 does the reasoning and planning, a separate action model turns images plus text into movement, and an on device version runs on the robot https://t.co/cNHH9JW1sx
@TherealJohnDoh@QuiverQuant filings dont show his total bill, but in 2021 he exercised nearly 23M options and his form 4 says about 10.6B of sales that round were solely to cover tax withholding https://t.co/izDcHAUQvx
@bryant_dan81996@SteveRattner ur not imagining it, BLS food at home went from 310.936 in jan 2025 to 321.500 in aug 2026, about 3.4 percent up, and gasoline is up around 30 percent. full numbers https://t.co/5a74a0c35G
@CWaldo53669@SteveRattner not sure where the 3.7 comes from but BLS all items CPI was 317.671 in jan 2025 and 334.980 in aug 2026, so about 5.4 percent since he took office. laid out here https://t.co/5a74a0c35G
@yashpanpaliya if selectors keep breaking on u, stagehand is worth a look, describe what u want in plain english, pass a zod schema and it comes back typed, runs in ur own chrome w just a model key https://t.co/0s5ejRdUaw
@krzysztofbury@akshay_pachaar read only db role as the real boundary is the right call ngl, hooks are a nice nudge but the role is what holds. same instinct for installs, these recipes pin exact versions and show the full change before touching files https://t.co/Y4XhSL8I8S
@wonjindev if u ever wanna test where embeddings beat keywords this is the tiny script i use, 6 docs and a question and it prints the closest https://t.co/2AePSyCtRR
@lumenxbt for tool names 100 percent, exact words matter there. where embeddings earned their keep for me was fuzzy questions w no shared words, and if the vectors are normalized dot product and cosine are literally the same number so comparing is cheap too