Siddhartha Reddy @sids

@dexhorthy Yeah but version control is the simplest way to share this with others who are also working on the same codebase. Everything else needs to make assumptions that may not hold with other teammates.

"AI safety" is like "speech safety": sounds good in the abstract, but it leaves out the most important part: who defines "safe"? If you only support "AI safety" when the arbiter agrees with you, then what you're really arguing for is control, not safety.

Quinn Slack @sqs

4 days ago

Every human has Freedom of Intelligence. Nobody & no gov't should infringe on our right to use available intelligence. Companies shouldn't be compelled to offer models freely, but they should be shunned if they engage in intelligence censorship, esp. if they lobby gov't for it.

11 8 110 16K 6

@DeepXP @claudeai I agree with that part. @linear does this very thoughtfully.

@guinnesschen This is great. But “Codex can orchestrate the handoff for you” world be even better if there was a button for it.

And they make halwa with this. Like, dude, why spoil the name of halwa!

Krish Ashok @krishashok

5 days ago

Consider the Bottle Gourd. If you had even the slightest bit of good taste, you'd probably consider it with a scowl, because its legendary watery blandness required coconut, chillies, tamarind and pure desi parental emotional blackmail to turn remotely palatable.

88 38 661 129K 89

Not only is Pi a great coding agent (my favourite, by far), it’s a surprisingly good SDK to build custom agents on. I built an internal slackbot around Pi and was surprised by how well it works. It’s become the de-facto L1 support agent at the company and provides the root cause for a good chunk of issues. It was really easy to have a custom structure (each Slack thread is a Pi session with channel-specific workspace & skills). We implemented a bubblewrap -based sandboxing solution that is working well for us, and this was possible because we could customise all the in-built tools.

Armin Ronacher ⇌ @mitsuhiko

6 days ago

It's so cool seeing people build stuff on pi. Makes me want to do scheduled office hours.

23 1 382 20K 20

A few additional thoughts on the Telegram ban in India: 1. Is this a permanent ban? It's unusual for the Indian government to issue a press release about the telegram ban: the precedence that I remember is when Chinese apps were banned in India in June 2020 including tiktok. A Section 69A order to ban something doesn't necessitate a press release, in fact it is protected by a secrecy clause, which should be held unconstitutional. So what prompted this press release? Is this a signal? Were they expecting a backlash and a PR puts a govt justification (as hollow as the Chinese app ban, which I supported) before the reaction. A BJP guy I spoke to the night of the Chinese app ban told me if they had their way, this would be permanent. So, is this ban permanent? As in, after things die down, will the date be extended? 2. This looks terrible for India and will impact us: It makes us look like a banana republic. Who in their right mind blocks an app used by hundreds of millions of users because of AN EXAM? How is this even legal, and is such a short-sighted-knee-jerk-uneducated-action-even-justifiable in a democracy? We talk about ease of doing business but for us it's a slogan. Which global business will want to operate in an environment of such regulatory uncertainty? We've seen some exit: toyota, citibank. This will play out across the world. It will definitely make it to late-night-comedy on US TV shows. We'll be a joke. Blocks telegram because of EXAMS? These are important signals that investors look at, and investors like going where things are predictable from a regulatory standpoint (unless they're playing a high stakes game), and not operating on whims. Look at the Crypto ban that pushed Indian crypto companies overseas. Look at the advisory on AI regulation that made founders question about building AI in India, before it was hurriedly rolled back. And this does look whimsical, unconstitutional and the idea of some guy with power who decided this is how things go. I mean, look at the US these days. Same thing. Varun Grover's "kabootar ki barfi" commentary comes to mind. At this point in time, as a nation, we need to project regulatory certainty and rule of law in a global economy where money doesn't know where to go. If the CEO of a company has to justify investing in India to her board, with the limited understanding that people have about India, the board is likely to push back because its their job to protect investor money. We're failing at this spectacularly, and I'm not saying this just wrt the Telegram ban. We're doing the opposite of what we ought to do. 3. This is precedence-setting. Once this goes through unchallenged, this will happen again and again and again. it happened with Internet shutdowns, and the Indian government got a lot of criticisms for years, for over 100 Internet shutdowns a year in the country. In some cases, especially Kashmir and Darjeeling, there were over 100 days of bans. Kashmir had bans regularly, for years. At that time we argued that this won't happen in a major city. App bans are easier: you block one, and people switch to an alternative, because as Indians, we adjust to government diktats. It hurts less than an entire Internet ban. IT Cell kicks in and says: - "you can't sacrifice using a messaging app for your country?" -"They didn't follow government orders, so this is right." - "there's lots of harmful content on telegram (pick anything from terror financing to port to copyright violation to app features)" - "It has harmful features like message editing that can lead to people being mislead that an exam paper has leaked." I'm surprised tech company founders haven't been given rewritten tweets on this issue btw. It's a thing and I'm surprised to see some friends of mine tweet a copypasted tweet recently, but I understand their compulsions: they're in a high risk business in a high risk country with little or no leverage. 4. Is this a smoke-test for Whatsapp? I'm wondering about how the decision went, and who took the final call. We know that it probably wasn't MEITY: for things like such bans, MEITY bows to Home Affairs. It's a BIG decision to take, given the hundreds of millions of users impacted. Did it go up to the PM? Ofc heads won't roll, like in case of the education minister, and that was demonstrably a shit-show that deserved a public sacking. this is smaller. This is kinda like the Indian Internet's demonetization moment, except there's no Paytm publishing ads in newspapers, and no payments industry celebrating a windfall, leeching growth off millions of people hanging off a cliff trying to get access to their own money. Media is already controlled and probably won't cover this well, and social media and YouTube is all there is, and it's tiny, and this is Telegram not Whatsapp. This makes me wonder if this is a smoke test as leverage against Whatsapp. Remember that the streaming services did not challenge the IT Rules in 2021 after what happened in case of Tandav. 5. Legality of the ban: legality of a ban is not a consideration anymore in India since the IT Rules challenges (they're largely unconstitutional) are stuck in courts, and in my opinion Indian courts are doing little about constitutional rights like freedom of expression, so that has given confidence to the government to create an infrastructure for censorship, and increasing censorship over the last five months. They're not responding to RTI's btw. We tried. Make no bones about this: - This act is disproportionate. There is legitimate speech that is being pre-censored with a ban on an app like this. - there is no clear cause-and-effect impact in terms of public order here. - The short term of the ban (so far) doesn't justify the scale of the impact, though that might be leverage in courts because by the time the hearing happens, the ban may be lifted so no one may actually consider challenging it. I'm of the opinion that except in some cases, our courts, including our Supreme Court, haven't upheld constitutionality and individual rights of citizens that is core to the functioning of a democracy. But that's a global trend, and this too shall pass. I think that when courts act to protect our rights, its an exception rather than the rule. Most of the time, like the SC did in case of the Right to Privacy (700+ days before a nine judge bench was founded), they just delay justice. 6. There are reports, and I don't understand this too well since I've not tracked network infrastructure for a while, that Internet landing station infra is being used to block Telegram. Ignore Durov's conspiracy theory claims of Whatsapp initiating this, because afaik this is a different Reliance (ADAG). Plus I don't think Whatsapp is that stupid or short sighted. But for those in networking, is the usage of landing stations to block an app unprecedented? Anyway, these are just some things I was thinking about so dumping it here.

there's a new word i'm hearing a lot in the most frontier-pushingest coding-agent builders: _program design_ for even the best agentic coders trying to maintain code quality, we've all seen it - you come up with something to build - you research the codebase, riff with the agent, align on what the end state looks like - you (or the agent) breaks it down into tasks for individual agents / context windows - you rip the implementation - the code works or is close to working - and it follows your spec to the letter but the code itself is still trash - poorly factored methods - leaky abstractions - tramp data - overloaded interfaces - try catch, useEffect, global variables everywhere I thought models would catch up, or that this wouldn't matter - that if we stayed in spec-land, understood the high-level architecture, and tokenmaxxed hard enough, we would be able to skip code review and just stay shipping doesn't seem to be working out that way I have seen agent-owned codebases spin up out of nothing... ...and I have seen them collapse into rubble within 6 months now there's something to be said about "skate where the puck is going"... ...and I can't tell you what tomorrows models will be capable of but I *can* tell you that *today*, models are mid-to-bad at program design you can solve some of this with memory / agents.md, but the scope of program design is massive. - entire companies have been built to help you implement it - books, classes, and professions have spun up around it are you building something to last? Or are you slinging more slop on the pile? anyways, thats the post, stay tuned for a fun announce tomorrow y'all 🙂

This is shocking, but unfortunately not surprising. So much for Digital India.

Techmeme @Techmeme

a week ago

India blocks Telegram until June 22 citing its use by organized "cheating rackets" to "defraud candidates" taking a national medical entrance examination (Reuters) (Visit Techmeme dot com for the link and full context!)

0 0 2 2K 1

India hasn't been the first in any of the large areas in the past. We got our internet access right two decades after every one else. We skipped most of the CDMA drama and went straight to GSM+3G after the world had figured out mobile telephone way earlier. We were late to payments, but figured out a way better way to clear payments than most others. We have been late to every new technological change, mostly because we've been loathe to invest large sums where failure is a given. Even our space and nuclear programs - where we have had to work indigenously with local research - we've taken a lot of time, to avoid major failures. The cost of failure is not just financial in India, it's the fear of shame that runs way too deep. New nice looking road? Oh look, not enough traffic on it, shame. That rocket didn't take off? Shame on us, we should have worked harder so it wouldn't fail the first time. Failed an exam? Shame. A gap year in your resume to try something out? Shame. This feeling is pervasive, but it's changing as we speak. As it does, India is investing into things, and we embrace failure a lot more these days. This will come from the next brand of companies that aren't afraid of failure, perhaps not so much from the older set of companies that have designed themselves to be cash flow giants. Not just IT, but in FMCG, etc. (Though in metals, we have seen massive risk being taken) There are companies that have transformed themselves through very large risk oriented projects - the biggest company in India is one. But I believe technological risk will be taken by startups and will be funded by people in the space who've seen what risk can produce, through successful exits in ESOPs or as founders from companies in otherwise boring spaces. We've lacked that capital really, because otherwise capital has been preserved in large families - risk taking is obviously stunted there as preservation takes a larger role in their future. Now that the first gen capital is more, expect some of this to happen. And no, I don't think the government should play this role (they can help by deregulating business) - it's private capital that should fund the next big thing. From the people that don't consider it shameful to fail.

There is an even more basic way in which they differ: we don’t build on top of the compiler output the way do on code, LLM generated or not. So even if LLMs generated deterministic code for a given prompt, we should still read it. At least inasmuch as it is going to impact the system design and affect all future changes.

@nixxin A bit of concentration (of capital, talent, demand) is necessary to reach capabilities that are worthwhile. I’d rather we have one good player than several mid ones.

People need to chill. This is likely just because of embeddings (vector) search, not an evil/genius idea.

NDTV @ndtv

2 weeks ago

Blinkit Returns Chocolate Results For Typing Gibberish Like Toddlers, Customers Call It 'Terrifying' ndtv.com/offbeat/blinki…

39 36 1K 323K 137

I hope not all GCCs go this way, but I fear that a lot of them will 😔

Techmeme @Techmeme

2 weeks ago

Opendoor says it is shutting down its India operations and laying off nearly 250 employees, replacing them with smaller, AI-enabled teams in the US (Moneycontrol) (Visit Techmeme dot com for the link and full context!)

4 6 52 13K 4

Apple showed how llms can be made really useful in human chat apps, even if it is in only small delightful ways. Meta completely dropped the ball here, treating WhatsApp only as a distribution channel to compete with other model labs. They are not thinking of the users at all.

signüll @signulll

2 weeks ago

even if siri does nothing more than become an llm layer on top of imessage, that’s still an incredibly powerful moat. you don’t need a revolutionary new product when you already own the place where billions of people communicate about every aspect of their lives.

37 13 387 17K 21

Good take on Anthropic’s automatic fallback from Fable 5 to Opus 4.8 and more importantly silently degrading its performance.

Drew Breunig @dbreunig

2 weeks ago

@tjl Nathan’s take is good: interconnects.ai/p/claude-fable…

2 0 8 1K 5

@lucasmeijer @_laurynas We’re giving the agent its own github user + making it add the user who triggered the turn that led to opening the PR as a co-author (+link to the link to the thread in the PR description).

There's some wild disconnect that people believe productivity is spending 6-figures on inference in a month. You know what happens if that remains true? We hire more software engineers and spend less on inference. "Loops" - or any other equiv - only work if compute is cheap.