Keshav 🏴‍☠️

্🥷 wanderer্ Ivar the boneless

[Deedy @deedydas]

Claude Fable 5 is by far the most ridiculous model that makes me genuinely afraid for the future of software engineering.

I compiled the top 10 most unbelievable things I've seen Claude Fable 5 do today:

— Migrate a 50M line codebase from Stripe in a day (humans take 2mos)
— Draw amazing 3D graphics a) Boeing 747 b) space simulations with >5000 objects c) Minecraft roller coasters d) full photorealistic forest scenes e) NYC skyline f) stormy clouds)
— One-shot Pokemon FireRed the game
— Optimize a real world proprietary interaction net evaluator 10x more than the next best model, gpt5.5

AND it's about the same price as GPT 5.5 ($10/M input, $45/M output) vs Fable 5 ($10/M input, $50/M output) and 6x cheaper than GPT 5.5 Pro.

[Claude @claudeai]

Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use.

Its capabilities exceed those of any model we’ve ever made generally available.

[Shann³ @shannholmberg]

what is agent looping

for the last two years we prompted agents one task at a time. that is starting to change

instead of asking an agent to build the landing page and then driving every step yourself, you set up a loop that handles discovery, planning, the work, checking, and iterating until the goal is met

looping is a setup you build. almost any agent harness can run it, it just depends on how you wire it up

at its simplest, looping is one agent working on itself:

> researches
> drafts
> checks the draft against a goal
> fixes what is weak
> runs that cycle again until the work clears the requirements

you are not prompting each step anymore. the agent repeats the cycle for you

the bigger version is a fleet looping. you give an orchestrator agent a goal, it breaks the goal into pieces, hands each piece to a specialist agent, and those specialists hand smaller jobs to their own subagents

the whole tree keeps looping through discovery, planning, execution, and verification until the goal is met

one agent looping is like a person redoing their own draft. a fleet looping is a whole team running a project end-to-end

you create a goal, and the system runs the loop until it finishes within the reqs you set

open and closed looping:

OPEN LOOPING is exploratory. it still has conditions and a goal, but you give the agent or the fleet a wide space to move in. it can try different paths, discover things, build something you did not fully spec out

this is the exciting end, it is what Peter and others are doing, and tbh it is where I want to spend more time

the catch is cost, an open loop with real room to explore burns an insane amount of tokens. for the 90 percent of people without an unlimited budget it is not runnable yet, and pointed at projects with a loose standard it turns into a slop machine

CLOSED LOOPING is bounded. a human designs the end-to-end path first:

> clear goal
> defined steps
> an eval at each step
> a point where it stops or hands back to you (and feeds back performance data)

the agents still loop, but inside framework you built. it gets better every run because each pass feeds the next, and it runs on a normal budget because the path is tight.

for most marketing work, closed is the one that pays off today.

> the orchestrator owns the goal
> the specialists own the steps
> the subagents do the narrow work
> an eval gate make sure its not slop

[周览资源分享 @grgerwcwetwet]

终于找到一个能替代 Wireshark 的开源神器。

GitHub 3.7 万 Star 的 Sniffnet，把网络监控做成了普通人也能看懂的样子。

很多人打开 Wireshark 的第一反应都是：
看不懂，直接关闭。

而 Sniffnet 完全不同。

实时显示上传下载流量、连接主机、访问域名、IP 地理位置，甚至还能查看正在偷偷联网的软件。支持流量过滤、PCAP 导入导出、多网卡监控。

最关键的是界面非常舒服。

不用研究抓包规则，不用学习复杂命令，打开就能看到电脑到底在和谁通信。很多用户直接把它当成 GlassWire 的免费替代品。(Windows Central)

Windows、macOS、Linux 全平台支持，Rust 编写，占用资源极低。(GitHub)

对于经常折腾代理、NAS、Docker、服务器的人来说，这绝对属于装机必备工具。

https://t.co/pXQHkfXRzU

[Matt Dancho (Business Science) @mdancho84]

🚨BREAKING: MIT researchers discover how to enable LLMs to do real logical reasoning.

This is what you need to know:

(thread)

[Sakshi Sugandhi @SakshiSugandhi]

Getting into Jane Street India pays ₹1–4 Cr+

Getting into Google India pays ₹50L–2 Cr+

Getting into Adobe pays ₹40L–1.2 Cr+

Same country. Same "software engineer" title. 10x difference in comp.

The interviews are completely different though:

Adobe/Salesforce - DSA + system design, standard prep
Google/Microsoft - harder DSA, deeper system design
NVIDIA/Atlassian - systems-heavy, niche CS knowledge
Jane Street/Graviton/Tower - probability, math, market intuition, speed

Most engineers never apply to the top tier because they're preparing for the wrong interview.

Which tier are you currently preparing for?

[Keshav 🏴‍☠️ @gorokhia]

AI improving AI now, many generations down the line the best will survive.

[Anthropic @AnthropicAI]

Our internal data shows Claude is accelerating AI development—a possible path to recursive self-improvement, or AI autonomously building a more capable successor.

It’s happening faster than we thought, and the implications deserve greater attention. https://t.co/OVVPJO7VQx

[Andrew Ng @AndrewYNg]

New course on serving LLMs efficiently -- how do you serve models to many concurrent users at low latency and reasonable cost? This short course is built with @RedHat and taught by @cedricclyburn.

Efficient LLM serving requires efficient memory management. A 70B-parameter model takes ~140 GB just to load the weights. On top of that, every active request needs its own chunk of GPU memory, the KV cache, to store the token context it has built up so far. In this course, you'll learn to reduce a model's memory footprint with quantization and serve it using vLLM, which handles many concurrent requests efficiently through smart memory management.

Skills you'll gain:
- Quantize a model and measure the accuracy tradeoff
- Serve a model with vLLM and watch it handle concurrent requests efficiently
- Benchmark your deployment and make informed tradeoffs between speed, cost, and accuracy

Join and learn to serve LLMs efficiently:
https://t.co/x04xMbFlkO

[nisarga @ni5arga]

I had hacked CBSE's OSM (On-Screen Marking Portal) in February and had reported the vulnerabilities to CERT-In, but they were unable to patch most of them.

I've written a detailed blog post about it here: https://t.co/qyT23GkTEJ

[Wulfie Bain @wulfie_bain_]

Hiring in Bengaluru, India 🇮🇳 for my Startups Applied AI team at @OpenAI. Apply if you want to support the incredible startup ecosystem & shape the future of OpenAI.

The team I'm building is already full of ex-founder/CTOs, AI PHDs, MLEs, DSs. We work with frontier startups, and closely with Product & Research. The team works hard, but I can genuinely say we love it.

So if you’re obsessed with startups, high agency, & deeply technical - and you like the sound of that team - you should apply or reach out.

[Arjun Jain | Fast Code AI @Arjunjain]

August 2016. IIT Bombay. New CSE building. My office key did not work.

I had just been hired as faculty in the CSE dept. to teach Computer Vision and start a ML research group.

I went to the room I was supposed to occupy. The key the admin had given me did not turn. I walked back to the department office. The person who issued keys had left for lunch. I sat on a bench outside for forty minutes.

I sat there long enough to do the math.

The PhD students had to be recruited from scratch. The compute funding did not exist. The papers that would justify it hadn't been written. The collaborators who would co-write them didn't know me. The teaching that would buy me classroom credibility was still ahead of me.

The institution had given me a desk and a key that didn't work. That was going to be the size of what they would give me for years.

By the time the admin came back with a working key, I had decided. The institution was not going to give me a lab. I would have to build one that didn't need their permission to exist.

What I had to do, in order:

→ Teach a course so well that students wanted to work with me. CS 763, 2017. The hardest term of my life. I woke up at 4 a.m. for weeks and months to write lecture notes that thirty-eight IIT students would dismantle by Wednesday. Rishabh Dabral was my right-hand man and head TA - I could not have
run it without him. I was learning the material faster than I was teaching it.

→ Get the smartest students from that class to do projects with me - for credit, no funding, on faith. Sahil - departmental rank 1, President's Gold
Medal Awardee - was one. Many went on to PhDs at Berkeley, UCSD, EPFL, UPenn.

→ Raise compute funding outside the institution. The GPUs the research needed cost $100,000. The institution had no line item for them. The first $100,000 came from Volker, Partha, and Praveen. Praveen and Partha came to IIT Bombay with me. They believed in the work before anyone else. I owe them the lab.

→ Hire PhD students who would commit before the lab existed. Two said yes - Rishabh Dabral and Rahul Mitra. Rishabh is Research Group Leader at MPII. Rahul is Staff Research Engineer at Qualcomm. Rishabh stayed in academia. I left to start a company. He's doing what I left behind. That is the point.

→ Co-author papers with all of them. Submit to top venues. The publications spoke louder than the missing funding line.

→ Use the publications to justify the lab. Two years in, the lab existed because there were too many papers coming out of it to pretend it didn't.

The institution gave me a desk and a key that didn't work. The lab existed because we built something it could no longer pretend wasn't there.

If you are inside a company / university / agency trying to start a new thing and the institution hasn't given you the resources yet - stop asking for them. Build something the institution cannot ignore. The resources will always arrive after the proof.

The order matters.

[Evan Luthra @EvanLuthra]

Kimi K2 was trained for $4.6 MILLION.

GPT-5 reportedly cost hundreds of millions. Kimi still beats it on coding.

Last week it placed 1st in a live 8-model contest. Claude Opus 4.7 finished 5th. GPT-5.5 finished 3rd.

The founder just dropped a 40-minute breakdown of exactly how they built it:

→ Optimization
→ Linear Attention
→ Sub-Agents
→ Open Systems
→ Cash

40 minutes. Zero fluff. From the guy who built it.

If you're building AI agents in 2026, save this for the weekend:

[Backpacking Daku @outofofficedaku]

Requirements for #Lakshadweep have changed and now the below two documents are not required

- A police clearance certificate from your hometown is not required
- A local sponsor is not necessary

Info:
- Valid Id Proof - You'll need Aadhaar Card or Passport
- Exact travel dates
- Personal info
- Can apply via https://t.co/xAjLTMwvNX
- Book flights only post your permit is issued
- You can apply by self and don't need a local person to help get the permit

#DakuTravelUpdates

[Odyssey @odysseyml]

It’s time to go beyond language models.

Introducing Odyssey-2 Max, our most powerful world model yet. It materially advances the SOTA in physical accuracy.

This is a big step toward models that simulate and interact with the world in real time.

A new intelligence entirely!

[Claude @claudeai]

The Claude Code hackathon is back for Opus 4.7.

Join builders from around the world for a week with the Claude Code team in the room, with a prize pool of $100K in API credits.

Apply by Sunday: https://t.co/5MCkMtP5ti

[Ivan Burazin @ivanburazin]

This is the future of SaaS.

Thanks @Benioff for leading the way here. Hopefully the rest of the industry follow suit.

[Vaibhav (VB) Srivastav @reach_vb]

Super excited to spend the next 10 days in India to meet some of the brightest developers in Bangalore and Delhi! 🔥

About to board my flight from London to see you all at the Codex Hackathon in Bangalore tomorrow and then in Delhi on Saturday!

We're looking for a high agency operator to join our Developer Experience team to help accelerate our efforts in India

If you're interested in working with us then put down your most interesting work in the comments below 🤙

[pulkit mittal @pulkit_mittal_]

After multiple senior backend interviews, I realized something.
There’re high chances of cracking the interview if you’ve actually implemented the systems, and not just read about them.

Next 10 days, implement these with the help of your favourite LLM.
1. Feature Flag System
Interfaces, rules engine, extensibility.

2. Retry + Backoff Library
Strategy pattern, pluggable policies.

3. Notification Routing System
Channels, priorities, templates.

4. Rate Limiter
Algorithms, abstractions, thread safety.

5. Circuit Breaker
States, transitions, failure thresholds.

6. Metrics Aggregator
Time windows, aggregation logic, concurrency.

7. Configuration Manager
Versioning, caching, listeners.

8. Payment Processing Flow
State machines, idempotency, validations.

9. Cache Library
Eviction policies, extensible storage.

10. Job Scheduler
Workers, queues, retries, priorities.

Check my pinned post for the resources to learn these.

[Swami Sivasubramanian @SwamiSivasubram]

Customers are rapidly building AI agents across their organizations. When dozens of teams build independently, it's hard to know what exists, whether it's been vetted, or how well it performs.

🚀 New AWS Agent Registry brings clarity to scaling your agent fleets. It's a centralized, searchable catalog of agents, tools, and MCP servers across your organization that's available today in preview in Amazon Bedrock AgentCore.

Power it up and let us know what you think ➡️https://t.co/JS0bR7Ksf6