Cindy Sridharan
@copyconstruct
Followers
41K
Following
33K
Media
1K
Statuses
12K
The IR can be something that can take human language prompts and convert it into something a bit more deterministic that LLMs can reliably parse in a repeatable manner. My hunch is that this interface would most likely be some minimal dialect of Python.
3
0
7
It's a little bonkers that we're giving LLMs instructions like it's a toddler which currently produces code that runs the entire gamut from "spot on" to "wtf are you doing. stop. stop. just STOP". For coding, IMO, there’s a need for an LLM-IR (intermediate representation).
1
0
10
Caught in the middle of a political battle, @BlackRock reaffirmed its focus is helping millions of Americans save for retirement. They warned that injecting politics from either side risks undermining financial performance.
prosperityretirementalliance.com
Millions of Ameircans depend on the security of their retirement savings to live with dignity after a lifetime of hard work. That’s why the Alliance for Prosperity and a Secure Retirement (APSR)...
7
3
46
The fact that this site is now 76% b o t s tells me that the old Twitter verification process, which had gatekeepers and was perhaps a touch too restrictive (and sure, it led to stupid clout games of why some Twitter famous personality wasn’t “verified”) worked better than this.
2
0
20
The @nytimes doesn’t report “news” anymore. Just “juicy collection of narratives” with click-baity headlines, which then unleashes a tsunami of hysteria against all kinds of targets. Mainstream media, with notable exceptions, is 1-2 years “behind” most trends and happenings.
1
1
25
RT @jburnmurdoch: But, plot twist:. The much-discussed contraction in entry-level tech hiring appears to have *reversed* in recent months.….
0
141
0
AI agent demos would be better if they did tasks like contest an insurance claim (tedious work) than make a reservation at a restaurant. When I go out to eat, I want to personally explore menus, check out photos etc. It’s part of the fun, and I don’t want AI automating fun.
3
13
113
One thing AI models are *really* a godsend for is … scripting with bash. Bash is so terrible in so many ways, yet it’s ubiquitous and often the best tool for many tasks. Having AI models write most of my bash with me just having to do minor edits is such a luxury.
2
5
111
As code *generation* becomes solely the purview of AI, code *review*, qualification etc becomes all the more important. I expect people who review code to have *more* subject matter expertise, not less, especially for critical production code involving money or confidentiality.
Q from a dev: "I'm a native Android engineer. My manager is pushing me to be more 'generalist' and do e.g. backend or frontend work as well. I am hesitant because being a specialist used to pay better.". My take: "fullstack" engineers will become more in-demand, esp with AI tools.
5
3
80
Personally, I don’t feel the need to join in with everyone saying how tacky Bezos’s wedding was. If you’re vaguely familiar with Indian oligarch weddings, Bezos’s wedding feels like a quiet affair in comparison. 😅. A price tag of “only” 66m is like Walmart wedding territory.
2
0
34
As widespread use of AI for “homework” becomes the norm, I wonder if Western countries will adopt the Asian approach of … having your grades entirely be determined by closed book, hand-written tests and exams, where you’re not allowed your phone, or calculators, or computers.
China shuts down AI tools during nationwide college exams
3
0
26
This might be one anecdote, but what a soul-crushing environment to work in, with no leadership, no engineering mission, no camaraderie between coworkers, no clear ownership … just isolation and disorientation. And to think of all the VCs who said go work for Twitter 2.0 …
2
5
143
RT @jamesacowling: The #1 rule of scalable systems is to avoid congestion collapse. This is when a system slows down, starts timing out, t….
0
48
0
2021 was 4 years ago, which is quite a long period of time, but it just doesn’t *feel* that long ago to me. I vividly remember 2017, 2018, 2019, and even 2020. But 2021 - 2025 has become something of a blur that I can’t really tell apart. 🙁.
9
2
116
RT @_joemag_: Recently, I've been discussing throttling with one of our teams. It's a super deep topic, probably deeper than it gets credit….
0
21
0
Massive, massive, absolutely massive #nokings protest on Market Street in SF happening right now. 🇺🇸❤️
0
3
22
Nice set of action items as well. - move toward “fail open” architecture.- not everything needs strong, instantaneous, global consistency. - don’t run your monitoring on the infra it’s monitoring . Funnily, I blogged about some of these very lessons back in 2019 (2nd image).
7
8
257
Full report for yesterday’s Google outage. - not feature flagging a new code path.- null pointer crashing a binary when reading empty fields from Spanner.- remediation causing a thundering herd as there was no exponential backoff.- which required manual throttling to recover from
17
148
2K
A lot of times, it’s hard to even know what’s the best design unless you’ve tried out different ideas. Often the best ideas happen *when* you’re in the middle of prototyping, and it becomes obvious another design works better. AI (as of now) can’t make such course corrections.
5
1
22
AI can vastly cut down the speed of prototyping, no doubt. Unless you work on a team where you design every single API, every interface right down to every class, method, error returned etc in a word document first, before writing any code, I don’t see how AI can do it all.
1
0
15
Am I the only person who needs to prototype a few API’s before it becomes obvious what the right design/layering/abstractions are?. I doubt (at least when working with complex, existing codebases) telling AI agents to implement design docs will produce the most optimal design.
13
2
112