🔬 Core LLM track: You care about:
- Model architectures
- Training efficiency
- Benchmark scores
- Research publications
⚙️ Applied LLM track: You care about:
- User experience
- Cost per inference
- Time to product launch
- Customer feedback
One isn't "better" — but they're completely different jobs.
Which resonates with you? Start from your somewhere, and i am sure you will be able to cut the clutter after some time, and able to identify where you belong in this food chain.
This morning was wild.
I spun up 10 background agents and they quietly shipped what would’ve taken me 3 full days even with @cursor_ai - all before lunch.
The velocity at which we’re pushing code to prod now is insane.
AI has made us faster, but it’s also made attention to detail non-negotiable.
Because if you stop thinking and start blindly accepting AI’s output, you’ll ship more bugs than features.
AI used to amaze me.
Today it scared me.
We’re heading toward a world where one engineer can deliver the output of an entire team - and the only differentiator left will be judgment, and how well you can catch what your AI assistants don’t.
Wild times to be building.
Al flow state in software engineering:
- use github copilot for exploring open source codbase, generating architecture diagrams and diving deep into unknown code.
- use vscode with copilot and cursor for doing devops/sre tasks where outputs generated are for on time scripting
- cursor with personal prompt library to build long yer software engineering projects
- lately have heard good things about kiro and claude code for subagents feature.
- gemini and perplexity pro for general research topics and debugging.
What's your mix of doing software engineering
We’re past the infinite-gross-margin era, which means scale is now the only way to make real money. software used to be about selling strings; now it’s about renting compute. In ten years there will be far more multi-trillion-dollar companies earning 5% net margins across mind-boggling scale than there will be firms still riding the 75% gross margins of the saas decade.
The lesson of this era is one of scale: tiny incremental gains across unfathomable scale, guarded by unassailable levels of capex. The best investors of the scaling era have really just made one big macro trade over and over. at ipo, analysts mis-modeled facebook’s terminal value by an order of magnitude.
You can’t model your way or iterate your priors out of that. selling zero-marginal-cost strings of code was an emergent feature of a very different path than the one we’re on today.
Thank you @narendramodi and @adgpi for ending this war in 3 days.
We have close relatives in border areas, and it was pretty much tensed up state we were in.
More so ever, now that it is over, we can focus on building things, doing business/work and working towards our dreams.
It is not easy to achieve all of this while balancing a befitting response, internal politics and time bound results. Ask any corporate leader, this is an excellent leadership outcome.
P.S: We are still worried about those who are near affected areas, and are following protocols.
@OPoncz@dragonfly_xyz
Is there a possibility of using databases/keyspaces to enforce multi-tenancy.
In redis, database are numbers, and range from 0 to 15. Is there a possibility to above and beyond in dragonfly.
@MarcJBrooker A position is set of roles, and these roles need to be transitioned depending on the context.
Excellent insight, and very clear article. Thanks, will use this.
@arpit_bhayani@PracSharma How do you manage edge locations(to reduce latency) ? I am assuming CDN to play some sort of role here. But CDN is not everywhere in India ?
@arpit_bhayani@PracSharma How do you manage multiple bit-rates w.r.t to different platforms (mobile, tv etc) ?
What is the level of redundancy in place. is it two time or n time redundant ?
How much lag is from live to actual cast in general. What is good user accepted lag ?
@Meta@narendramodi@AshwiniVaishnaw
Kindly look into this, it may give @Meta an undue advantage, and I fear minors and our parents will suffer from this.
In fact, an LLM on networked data needs consensus of all who are involved in the network.
@Meta What is the logic for enabling LLM search in whatsapp.
Earlier you could not index conversations, but now you have a way in which you can index private communications of people.
How we can trust that no one is watching, peeking and analyzing on this ?
@Meta Most of my family/friends is on whatsapp, and I fear that even if I do not opt, but one of my close friend opt, then all of my conversation is gonna index.
This feature will be accepted cause your demographics has large number of people who does not know about security.