Pinned Tweet
Two execution traces can look completely different even when nothing important changed.
Timestamps, random IDs, and event ordering create noise that makes real changes hard to spot.
Built TraceDiff: feed it two traces, get back only what actually changed.
Demo Link Below
there are over 6000 paper on agentic memory on arXiv , how to decide which one to read ? there should be a some kind of rating system so that i could judge those paper 🙂
When i am reading these papers , each minute i encounter some new terms that i dont know about 🙂 , i was reading how mem0 cut its vector search latency by 70x .. now here they are talking about HNSW indexes i dont know anything about it
maaz retweeted
We’ve raised $1.3M from YC & Afore Capital so you never have to type or speak again.
Typing is slow. Talking into your computer is awkward. Subvocal fixes both
Our glasses give you the speed of speech without you making a sound or touching a keyboard.
It tracks your internal speech muscles with non-contact sensing and decodes silent, almost invisible mouthing into text at more than 200 words/min.
Pre-orders are now live at:
subvocal.co/preorder?utm_sou…
Two horrendous case in a day , what was going on in his brain to do such an act .. how was he desensitized to this extent .. they should do a through investigation to find out what led him to do this
If I were building something, I'd definitely take these things into consideration so that no one could drain the free credits
we lost $4k overnight 😭
woke up to 5.5k+ signups thinking we FINALLY went viral
turns out someone made thousands of fake accounts to drain our free AI credits.
even email verification didn't stop them lol
and what did they use all those credits for?? a WHOLE short drama series
and the part that actually pisses me off?
it was REALLY good
how am i even supposed to be mad at this lmaoo
maaz retweeted
i built weft, a cloud coding agent, to learn what it takes to make one trustworthy.
point it at a github repo and ask a question or request a change. it searches the code with grep and glob, edits in an isolated docker sandbox, reruns your tests in a clean one, and opens the PR.
I don’t think agentic memory is solved. Many systems compress or summarize memory once it overflows the context window. That compression is lossy, and since future queries are unknown, information discarded at write time may be needed later. Storing everything avoids this but shifts the problem to retrieval: at scale, recall, latency, and cost degrade, and stale or contradictory memories pile up. As personal agents and robots become common, they will need long term memory that scales .
maaz retweeted
Starting today, all Databuddy SDK versions ship with SRI hashes and versioned scripts, what does this mean?
if our upstream script gets hijacked, your website remains safe, the script won't load
While extremely unlikely in the first place, I ask everyone to update their SDK!
