NOT an MBA.

Joined August 2007
386 Photos and videos
Mitesh B Ashar retweeted
I was using Fable to improve security for my apps and when it suddenly stopped working. It's only on twitter that I found out that it has been stopped for people who are not US nationals, by the US government. Here's how I'm seeing this: 1. Fable comes with security safeguards to prevent misuse, and can be used to plug security loopholes. I've seen it identify and plug security issues in my apps. The US government refusing to let non-US nationals protect themselves is deeply problematic in an interconnected internet world, especially when these issues will impact US customers of foreign apps and websites. It doesn't make sense. This hurts everyone by making all systems vulnerable. 2. Yesterday Anthropic announced a partnership with TCS in India as a commercial deal. Today Fable is banned for Indians. During the AI summit they had announced a partnership with Infosys. The fact is that these are just commercial deals but the Indian companies have no strategic leverage here. Both companies run critical systems in India: Infosys runs finnacle. TCS runs passport sewa etc, all with critical personal data at stake. They had zero access to Project Glasswing when it launched to secure systems with Mythos, while some US companies got access. This proves that for Anthropic, Indian partnerships are just about the money. To call such partnerships strategic is hogwash. I think Anthropics leadership and policy teams in India have much to answer for. I would request @nishikant_dubey to take up this bipartisan issue up in the standing committee on IT since MEITY is quite toothless. 3. We've seen this story play out before: in President Trump's last term, the US govt momentarily stopped Android security updates for China. Today China has its own Huawei OS. Where is IndusOS? We need to adopt open source tech because geopolitics is increasingly win-lose, and the win-win era is behind us. It's sad but this is where we are today. We need to invest in research, and building a culture that supports research. The Atal Tinkering Labs are a great starting point and we need to build on this, at a university level. Allow universities to monetize research and professors and students to start companies with university support. Let professors consult. Build a marketplace that rewards innovation and expertise. 4. CERT-IN, RBI and other agencies holding meetings regarding mythos was farcical. Its proof that our cybersecurity agencies are out of their depth, and doing a checkbox exercise to show the PM that they're doing something when they can't, because this is about Zero Day vulnerabilities, not predictable cybersecurity issues. We need a cybersecurity strategy overhaul. Basically fire bureaucratic-mindset people doing farcical compliance at CERT-IN and get technocrats with actual cybersecurity understanding. Hire for competence, not loyalty. Hire for competence not badges (definitely not the clueless famous IIT professor we know who does committe hopping). 5. Time to start ignoring Nandan Nilekani's ignorant comments on what India should do in AI. We need to focus on hardware, start working on small language models and get people who know AI to drive policy. As history has shown us, Vishal Sikka was right, Narayan Murthy and Nandan Nilekani were not. We have a long way to go. The AI Summit was great for increasing diffusion, and we need more of that, but not just that. The IndiaAI mission needs to speed up and become a mission critical project. Cc @narendramodi. Hire better people please. Choose open source. Build a long term strategy.
19
92
279
35,985
To some extent, I can digest the idea that the government feels threatened for the nation. But it sounds absurdly funny to assume that US populace comprises of ZERO adversaries, whatsoever!
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Claude models is not affected. We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible. Read our full statement: anthropic.com/news/fable-myt…
39
Mitesh B Ashar retweeted
I cancelled my $10/mo Calendly subscription and vibe coded my own with Fable for $12,000
466
613
19,260
965,068
Mitesh B Ashar retweeted
My friend's MRI came on a CD we couldn't open, filled with 500 cryptic images. I built Dr.MRI.AI to fix that. An AI-powered, privacy-first tool using #Gemma4 to turn DICOM data into clear, clickable reports. Read the story: rkrants.blogspot.com/2026/05…
1
1
1
110
> What’s very surprising though is that using no context files is also cheaper and more efficient in their benchmarks! At first, I was surprised reading that it's cheaper. But the later explanation that AGENTS.md instructions resulting in increased tool calls explains it. Human-written files being better makes a lot of sense. I try to limit the content to very high-level and high-value instructions that are not easily derivable from the codebase and can impact the workflow to narrow down paths.
80
Mitesh B Ashar retweeted
The most important skill of the AI age will be coordination. You won't lose to an AI or a human with AI. You'll lose to a tight-knit well coordinated group of humans with AI. Roving bands of attention and inference hunting the solitary and the lumbering alike.
3
5
17
903
Folks learning AI from me have been asking me to share a list of people to follow on X, meant for non-technical and semi-technical people. I thought I'd rather put that list here than in direct messages. @mvanhorn @mattpocockuk @geoffreyhinton @badlogicgames @demishassabis @deedydas @theo @addyosmani @championswimmer @enjoyingthewind @claude_code @FredKSchott @GeoffreyHuntley @devagrawal09 @aparnadhinak @Vtrivedy10 @natebjones @rohit4verse @reach_vb @kieranklaassen @trq212 @dexhorthy @GergelyOrosz @gregisenberg @LLMJunky @AndrewYNg @thdxr @kitlangton @gkcs_
2
1
4
1,712
Mitesh B Ashar retweeted
Yeah let's worry about replacement fertility because we have already maximized the health, education and productivity of the people we do have.
5
18
188
10,229
Mitesh B Ashar retweeted
New day, new pwn 🫠 “Certified secure” lore just got funnier. Reports said Coempt gave CBSE cybersecurity certificates linked to OneX/BPUT/pre-production context not clearly CBSE’s actual OnMark production system. So I looked at OneX too, and guess what? Critical vulnerability found. Direct super admin access without OTP, verification, approval flow, hardcoded master password, brute force, credential stuffing, or social engineering. Just broken access control at the worst possible privilege level. Reported responsibly, and CERT-In along with the concerned team moved fast to get it patched. a certificate from some other scope or environment does not mean your real production ecosystem is secure. Audit scope matters. Production security matters more. #CBSE #OSM #ONMARK #COEMPT #CYBERSECURITY
2
12
84
4,549
Mitesh B Ashar retweeted
Got a bunch of replies saying something like this "ECI had an EVM hacking challenge which nobody did" but it's incorrect. Disclaimer first: 1. I do not say (and do not think) that past elections have been tampered with by rigging EVMs. Like I've mentioned same parties have supported and opposed the same EVMs depending on when/where they are/were in power or opposition mainly for FUD. 2. I do believe there should be an open audit of the EVMs. Now to the reply for this notion about "open hacking challenge" 1. An audit and a hacking challenge are very different. An audit may include red teaming, like an open hacking bounty/challenge but an audit is much more as it allows open access to the auditors to the internals that can then be verified. I've done a bunch of audits for HW security, and also a lot of my work over decades have been subjected to audits by external parties. 2. ECI's challenge was NOT a "hacking challenge"/"hackathon". It was a challenge to prove 2 specific statements about vote manipulation. 3. ECI did not give much access to the machines at all. Only enough, without any access to documents or code or hw or any allowance to change anything. The access was limited to only the allegations of button press combos and wireless access if possible as claimed by certain parties 4. ECI did not allow just about anyone to do this. This was limited to only political parties (and within those, only the ones that participated in the elections around that time). So only vested interests were still allowed anything at all, most of whom did not bring any decent experts, and even if they did, they couldn't have done anything given above limitations (plus a time limit of 4 hours) Again, this does not prove ECI was bad/in cahoots/was doing something nefarious. They did enough to disprove the allegations of certain parties about the specific "hacks" they were claiming have occurred. But it is not anywhere close to an open hacking challenge, and definitely not an open audit which is required.
Replying to @shantanugoel
Maybe you missed a huge bounty set by ECI for folks who can 'hack'.
2
5
29
4,137
Mitesh B Ashar retweeted
See, I have never believed that the divine purpose of human existence is to open some 3-pane text editor with weirdly colored text and hit random keyboard shortcuts all day. Even though I love building apps, and I love the flow state that comes with writing code in an IDE - I have no qualms about the fact that there is nothing to "romanticise" about the IDEs. A tech company CEO (or even many product managers) are after all "prompt engineering" their tech teams to materialsie a reality that they have already imagined. If you can shout at your computer and it conjures up an app, and it actually works, and your users actually like it and use it - so be it. Your end users never cared about how the developer felt when they in their flow state in their IDEs when making that app. But all that said, these statements from Claude team, at least today, feel extreme disrespectful to any and all users of their products. Because the quality and the competence of these agentic systems are just not there. They might be, in few months/years for sure, but today they are not. So when you do not care about the output they generate and only write loops that write more loops that generate prompts, yes sure some code is generated. "Something" comes out of it, but it is the ugly mess that Claude Code is. I implore anyone who has not used anything other than Claude Code, to once, just for trying out, use OpenCode or Pi or Amp or Goose. Just see the difference. Just see how many bugs you're literally living and breating on a daily basis. And they continue to pile on more into it.
45
21
392
73,498
Mitesh B Ashar retweeted
As someone who worked on lowest annals of hw security for most of his career, evm attacks are possible. No HW/SW is completely immune to them. But what this guy says here is just a gibberish of random words (I wouldn't even call them buzzwords because they are the most mundane ones, used to wow maybe the most tech-illiterates) Irrespective of the language used (although python is most unlikely to be used here), the key aspect is how one bypasses the code and system integrity checks, not that by changing the code they can alter the behaviour. That by changing the code one can do something is no-shit-sherlock domain. Any demonstrated EVM attacks require much more invasive supply chain attacks than what was required in the past by simple booth capturing. And with VVPAT, these attacks go down the drain pretty quickly as well. One would notice how any party in power is in support of EVMs and the ones outside are in opposition. Congress supported it in the 2000s with BJP questioning it. Since 2014, both of them reversed their stances according to their positions of power. The actual outrage from most such is to save-face spread FUD so they can turn people towards themselves, not to actually fool proof the voting systems. All that said, I do believe EVMs should be openly audited. Better that their flaws are identified and fixed, than having exploits being carried out by a select few.
Gen Z Cockroach: "We need to get back to Ballot Papers. I'm an IT Professional. -> I can HACK EVM easily by 3 codes. It is coded by Python Full Stack"😭 Drug dealers are definitely doing overtime around Jantar Mantar🤣
13
8
70
13,431
Mitesh B Ashar retweeted
Bio says capitalism. Candidate wants to do capitalism and suddenly they have a problem.
Replying to @sahilk
Current. CTC. Does. Matter. If you want to double your pay, you need to justify it. Simple.
13
51
869
22,506
Mitesh B Ashar retweeted
Wow!
Built a pottery app today, where your real hands throw virtual clay 🏺 Hand tracking > clay deformation > real-time physics No controller. No stylus. Just hands. Everything built within @omma_ai threejs Model of the puppet hand: credit to LiamVandeWouwer (SketchFab)
2
4
50
10,230
Here is a major context engineering bug in Claude Code that bloats output token usage. Have you noticed the below? Tell us about it! Observation: Claude Code agents often repeat verbatim file contents in their subagent prompts. Why that is a problem: It could have provided file paths for subagents to read. Instead it wastes output tokens, which are much more expensive than input tokens! Imagine that happening for "n" parallel subagents being invoked. Why this may be happening: - Deliberate instructions in system prompt to do so: I cannot imagine a reason for them to do this, except higher billing. - Bad judgement by models: Which could happen due to confusing instructions in the system prompt, or due to lack of guardrails for this scenario of agent invocations.
3
2
92
This fabulously written article by @mitsuhiko about the perils of humanizing AI gives me a very MMORPG vibe about our landscape! A lot of the discourse around the word is opinionated expressions, which is totally fine. The important bit getting lost in the discourse is at the end of the article - the boundaries and the responsibility.
More musings after some people got upset about the word clanker. lucumr.pocoo.org/2026/5/26/c…
51
Mitesh B Ashar retweeted
The best ai fanfic I've seen so far.
Anthropic onboarding day: Michael Scott introducing Karpathy like he just signed Wemby in free agency.
4
4
130
26,503
Mitesh B Ashar retweeted
This also Soros agent? Unbelievable scenes. From a major Indian city.
Nothing to see here. Just a ‘cockroach’ crawling on the road and begging for something as basic as WATER.
10
164
1,773
20,789
Mitesh B Ashar retweeted
This is why data centers are coming to India because they are facing this push back back home. Here the people and the government don't care if the water is that color or poisonous as long as they can hate and discriminate against minorities.
May 21
AOC: This is what drinking water in Georgia looks like after Meta began data center construction in the community.
65
1,934
6,893
126,244