i think about this often
It's actually crazy how early I was with a lot of things. I'm not the type to really make predictions, but instead, i kind of just "do" the thing that is going to become big very soon. Like I'm really deliberate with it. It is the best way I can communicate what I know. By play
16
1,060
16,686
if you want better thoughts and ideas, you need to be rawdogging sources more instead of using ai for it. use ai for filtering and curation instead... you also need more contact with reality via experiments.
some excerpts from my blog post
sankalp.bearblog.dev/on-idea-genera…
some excerpts from my blog post
sankalp.bearblog.dev/on-idea-genera…
3
2
42
i might become grok bot fan (because it works in sync with x chat (it can go long back into history!), can read through posts and lists, set up situation monitors etc.)
1
0
10
me when my mind coined the sentence "mech int is like astrology for ai people"
can i say something without everyone getting mad... i often see some mechanistic interpretability experiments here and people getting excited about them without really understanding what's going on and sometimes it feels like mech int is like astrology for ai people lol
6
0
83
these quote tweets suddenly feel more visceral now. need to sign apology form to both anthropic and openai if i criticized the holding back of models.
Introducing Project Glasswing: an urgent initiative to help secure the world’s most critical software.
It’s powered by our newest frontier model, Claude Mythos Preview, which can find software vulnerabilities better than all but the most skilled humans.
anthropic.com/glasswing
1
1
27
quick coverage of posts and relevant articles on the CVEs posted yesterday
trackingsingularity.com/e/php2shell/
trackingsingularity.com/e/php2shell/
0
2
30
openclaw -> cringe
hermes -> nice greek aura
grok bot -> cute name
dots -> meh
claude duet ->
hermes -> nice greek aura
grok bot -> cute name
dots -> meh
claude duet ->
We’re opening a research preview of Claude Duet, a new app where Claude works as a proactive agent.
It comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf.
15
21
743
curious people and people interested in llm evals, benchmarks, high data circles, please wake up. my blog post "SWE-bench Multimodal v2 is not multimodal" is up now.
sankalp.bearblog.dev/swe_multimodal/
sankalp.bearblog.dev/swe_multimodal/
39
25
329
good, i know the perfect guy for this position
“We regret to inform you that we won’t be moving forward your application forward. There appears to be a misunderstanding of required qualifications. As such, “two decades in command of Ultramarines” is not sufficient. We wish you luck in your search.” job-boards.greenhouse.io/anthropic/jobs…
1
2
119
me to my past self 3 days later after realising i am still working on the blog i thought i would finish quickly
3
0
43
they are still releasing this stuff on end of week😭. i know it's less concerning but yeah.
2. June 26: A GPT-5.6 Sol family model was asked to deploy cryptocurrency smart contracts. It found and used a funded wallet that the user had not supplied. It derived keys from public developer recovery phrases, checked 362 addresses, and found one with enough funds to pay for it was just an Ethereum test network.
1
2
19
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had
with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works
already-clean.goodfire.com
9
11
444
on anthropic making their own version control system scion with claude
@dejavucoder This is like an old problem to solve -- meta, goog, and even amzn have their own internal version control systems to handle the share volume of commits. These system have been working extremely well and 100x better than git.
0
1
19
you only realise this after you develop the eyes. in the fromsoftware game bloodborne, you can gain insight to gain "eyes" and then you start seeing more of the creatures in the world.
1
0
2
i dont know if ivan burazin writes all his tweets but having read his tweets over last 6 months around rl infra / sandbox infra stuff, bro is this person (or ofc has a person hired).
1
0
2
@henloitsjoyce yeah she's really cool. i got to know her via YVES (when they posted the new version of pop girl
youtu.be/e_njstiXD40?si…
youtu.be/e_njstiXD40?si…
0
0
1
