[ 66151680.000 ] posts.x: Full talk on training a search sub-agent with RL, and the speed and cost gains that come with it:  |   [ 66153540.000 ] posts.x: @hypeapps This kind of time tracking is quite important. Until you’ve logged how long projects (and, if you’re as easily distracted as I am…  |   [ 66153720.000 ] posts.x: @Mappletons I’ve been redesigning my personal site this past week and I’ve studied your site extensively as a great example. I really appreciate your…  |   [ 66153780.000 ] posts.x: @Mappletons The redesign isn’t live yet, BTW …  |   [ 66159420.000 ] posts.x: Full talk on why there's no single right chunk size and how multiscale indexing with RRF closes the gap:  |   [ 66159420.000 ] posts.x: A fixed chunk size is a bet on queries you haven't seen yet. @yuvalinthedeep, Sr. Developer Advocate at AI21, tests that bet in "Stop Chunking Like…  |   [ 66166140.000 ] posts.x: @hypeapps @toggl Is this because you’re straddling multiple agent sessions for multiple projects simultaneously?  |   [ 66166560.000 ] posts.x: @hypeapps @toggl You could still get pretty close. Agent sessions are stored on disk. Subtract the long gaps between your inputs to the session…  |   [ 66168660.000 ] posts.x: Full talk on why llms.txt isn't enough and what actually makes a website agent-ready:  |   [ 66168660.000 ] posts.x: Almost half of the websites in one study already publish an llms.txt file for agents to read, but almost none of the agents actually use it…  |   [ 66169620.000 ] posts.x: Full talk on why AI cluster networks need a receiver-driven, message-based protocol instead of TCP:  |   [ 66169620.000 ] posts.x: Most AI clusters still tune their networks for giant weight transfers, but the workloads pushing performance limits now are tiny messages: a KV cache…  |   [ 66179400.000 ] posts.x: Full talk on the harness layers, the files-vs-databases tradeoff, and context rot:  |   [ 66179400.000 ] posts.x: Most of what makes an AI agent reliable has nothing to do with the model itself. In "Total Recall: Agent Memory and Harness Engineering,"…  |  

[ 66151680.000 ] posts.x: Full talk on training a search sub-agent with RL, and the speed and cost gains that come with it:  |   [ 66153540.000 ] posts.x: @hypeapps This kind of time tracking is quite important. Until you’ve logged how long projects (and, if you’re as easily distracted as I am…  |   [ 66153720.000 ] posts.x: @Mappletons I’ve been redesigning my personal site this past week and I’ve studied your site extensively as a great example. I really appreciate your…  |   [ 66153780.000 ] posts.x: @Mappletons The redesign isn’t live yet, BTW …  |   [ 66159420.000 ] posts.x: Full talk on why there's no single right chunk size and how multiscale indexing with RRF closes the gap:  |   [ 66159420.000 ] posts.x: A fixed chunk size is a bet on queries you haven't seen yet. @yuvalinthedeep, Sr. Developer Advocate at AI21, tests that bet in "Stop Chunking Like…  |   [ 66166140.000 ] posts.x: @hypeapps @toggl Is this because you’re straddling multiple agent sessions for multiple projects simultaneously?  |   [ 66166560.000 ] posts.x: @hypeapps @toggl You could still get pretty close. Agent sessions are stored on disk. Subtract the long gaps between your inputs to the session…  |   [ 66168660.000 ] posts.x: Full talk on why llms.txt isn't enough and what actually makes a website agent-ready:  |   [ 66168660.000 ] posts.x: Almost half of the websites in one study already publish an llms.txt file for agents to read, but almost none of the agents actually use it…  |   [ 66169620.000 ] posts.x: Full talk on why AI cluster networks need a receiver-driven, message-based protocol instead of TCP:  |   [ 66169620.000 ] posts.x: Most AI clusters still tune their networks for giant weight transfers, but the workloads pushing performance limits now are tiny messages: a KV cache…  |   [ 66179400.000 ] posts.x: Full talk on the harness layers, the files-vs-databases tradeoff, and context rot:  |   [ 66179400.000 ] posts.x: Most of what makes an AI agent reliable has nothing to do with the model itself. In "Total Recall: Agent Memory and Harness Engineering,"…  |