@obeserverji
Account based inIndia!
About this account
- Account based in
- India
- Connected via
- India App Store
X says this location may be affected by a proxy or VPN.
Account-level information from X, not a live location or the device used for a specific post.
suffer my typos and bad trading takes #datakhore(tm). generating and discarding useless signals and metrics since 2015
New Delhi, India
Joined October 2016
- Tweets2.7K
- Following908
- Followers144
- Likes3.5K
2cents retweeted
I like this article!
The problem is not for some people to care and work on catastrophic risks (we actually need much more research into it), it’s when company leaders and policymakers use fears of them to distract and hide the current challenges and solutions!
quillette.com/2026/09/24/the…
2cents retweeted
At this auspicious hour of Krishna-pradurbhava, I present BhagavataVaNi. A comprehensive app for Srimad Bhagavatam. 10 Indian Scripts, searchable, indexed, with Voice over and Karaoke. Your handheld companion for Bhagavatam.
Krishnarpanamastu.
Link in the first reply.
2cents retweeted
Heard that some frontier models are basically a 48-layer transformer looped twice (48L x 2).
Now we are introducing DeepLoop: Depth Scaling for Looped Transformers (arxiv.org/abs/2607.13491),
making the loop transformer stable and scalable!
The Future Is Domain-Specific Agents - Justin Schroeder, StandardAgents youtu.be/spNAUEgq_A8?si=XTOa… via @YouTube
2cents retweeted
Samsung runs 68 training centers in India, TSMC built talent programs years before opening factories abroad.
Dixon is copying that same approach at BITS Pilani. It picked teachers, wrote courses, and sits on governing council. Students study displays, optics, robotics, precision engineering. Second program at BITS Hyderabad focuses on product design. Third at Plaksha University covers scientific research. 3 different schools, all customized by Dixon.
Companies that build their own worker pipelines grow faster because they do not wait for someone else to train their people.
Dixon decided to fix this instead of waiting for government programs. It launched training center at BITS Pilani where Dixon picks teachers and designs courses.
Students learn about displays, cameras, optics, robotics, and precision manufacturing, all tied to what Dixon factories need. Same program runs at Plaksha in Chandigarh. When companies design training, workers show up ready from day 1
Dixon used to only assemble phones and TVs. Now it makes display screens, camera parts, and batteries. Revenue up 25% from last year But profit margins fell because simple assembly grew faster than higher-value work.
Making displays and camera components improves profits But those products need engineers trained in optics and cleanroom manufacturing. Indian colleges do not teach those skills at needed volume.
Dixon BITS Pilani program exists because without right talent, backward integration stays on paper and margins stay thin
L&T does same. well now i feel good University Industry Partnership is must needed
Manufacturing making deeper in roads not just in industry but also in education -
"Dixon has now launched a center of excellence in BITS Pilani, which from August is going to be rolling out an M Tech program, specifically in these areas of display, optics, artificial intelligence, robotics, humanoids, tools and dyes, precision engineering.
And the courses have been curated. The faculty has been selected all by us. There's a governing council, which I lead along with Vice Chancellor of BITS Pilani. The same thing is going to be replicated, and that's for product designing at BITS Pilani Hyderabad campus.
And for the larger initiative on the scientific side, the same M.Tech program is being launched at Plaksha University Chandigarh Mohali. So the whole idea is deepen the partnership with possibly the best globally, bring in foreign talent or nurture Indian talent to acquire that skill set. Now it's going to take time, but that is the building block one is trying to put together."
No reco , src – Dixon Q1 concall
try Autolith - a live Common Lisp agent image you (and the model) can redefine on the fly — Emacs-level extensibility, shared REPL, automatic recovery if it goes sideways, checkpoints, and actual low RAM use.
Not another wrapper.
github.com/lambda-symbolics/…
What makes Autolith unique as a harness (long list):
- Self-modifiable, self-introspectable live image
- Both user and model can redefine essentially any part of the running system
- Much closer to Emacs-ish live extensibility than a conventional agent CLI
- Automatic recovery-image boot if the active image gets lobotomized
- Recovery image can diagnose failures and roll back to an earlier image generation
- Built-in triage tools often avoid needing a full rollback
- Native RLM support as a first-class toolset
- RLM environments can manipulate external context and recursively invoke inference
- Full Common Lisp programming environment shared by user and model
- Prompt input is itself a Lisp REPL
- Plain text input is just syntactic sugar for `(prompt "...")` lol
- Tools are ordinary Lisp functions
- Users can invoke the same tools directly
- User-side tool calls appear in the shared log, so the model sees what you did
- Very useful for interactive debugging
- Long-running REPLs instead of disposable shell sessions
- REPL state can be checkpointed
- Failed experiments can be rolled back to earlier checkpoints
- REPL images can be dumped and reused later
- Native builds for FreeBSD, OpenBSD, and NetBSD
- macOS and Linux on both x86-64 and ARM64
- One of the few (only?) mature-ish agent harnesses written in Common Lisp
- Follows XDG conventions
- No fullscreen TUI
- Preserves normal terminal scrollback
- Uses standard terminal colors, with no assumptions beyond xterm-256
- Human-readable, templatable Org Mode system prompt
- Relatively low RAM usage under normal workloads
- Fairly small codebase for the features
- Owns the model loop directly instead of wrapping or launching another agent CLI
- Multiple running Autolith sessions can be inspected and controlled via localgroup
- Sessions can be attached, detached, paused, messaged, and killed
- Native Skills with request-local instruction loading
- Native MCP support for tools, resources, and prompts
- Persistent memories at both workspace and global scope
- Ranked request-local memory recall
- Agendas are first-class persistent state
- Agendas can link back to memories
- Fast in-process repository search via fff (thanks @neogoose_btw for making it!)
- Path, glob, exact, regex, fuzzy, and multi-pattern search modes
- Worker images can be modified independently
- Pristine images can be retained for side-by-side comparison
- Ships matching SBCL source for implementation-level introspection
- The agent can inspect below the harness into the Lisp implementation itself
- Durable self-modifications have replayable reconstruction artifacts
- Self-modification can be disabled while retaining introspection
- Each image generation has reconstruction artifacts
2cents retweeted
A post of ours on vendor lock in and LLMs. We don't like the growing trend of AI companies quietly hiding your data while stripping away your control. We think that's bad for users and bad for the ecosystem. Here's are our thoughts: earendil.com/posts/session-p…
2cents retweeted
🚨 Before India's private space boom, there was one early bet. 🇮🇳👇
When Mukesh Bansal invested in Skyroot Aerospace in 2018, India's private space sector was yet to be opened to private players.
There was no proven private launch ecosystem, no space-tech unicorns, and no roadmap for commercial orbital launches.
Eight years later, Skyroot's Vikram-1 has become India's first privately developed orbital rocket, proving that some of the biggest breakthroughs begin with investors willing to back the impossible before everyone else. 🇮🇳
