Abhishek Malik
1.8K posts
Abhishek Malik
@abhimskywalker
Building products | Hungry and foolish | Ex: Product @Paytm, Cofounder-CTO @BalanceAppHQ, DataScience @Citi, Economist @IITK
New Delhi, India Katılım Mayıs 2009
1.1K Takip Edilen605 Takipçiler
Announcing the Audiobooks Creator Challenge.
Create.
Publish.
Win.
Win AirPods Max & credits to publish your full book.
English
@thsottiaux - iPhone app please
- Please allow compaction like cli to be user initiated…
- loading takes a lot of time… when conversations go long… auto compaction is not as good as the cli version
- no teleport session to cloud or a worktree 😓
English
if you're vibe coding or building over the holidays, i want to gift one of you a 6 month subscription of claude pro to support <3
just drop a comment below. merry christmas!
English
My year in code with Cursor!
Well only my past 6 months since I got a team account, so I'm not making the top of any leaderboards haha.
Whoever responds with the most tokens used building, I'll send you a bunch of Cursor credits 🫡
English
@tom_doerr Yes. Cursor with gpt-5-high and it’s fast variant have been most error free relatively…
Claude 4.5 for speed for smaller scoped edits
But gpt-5-high for larger scopes have almost always been better in getting to target without much regression bugs and misses.
English
@JuanRezzio - Adding a kanban board sorts todo for better transparency for both AI and human
- More EM like back and forth to go from ambiguous to clear specs
- Auto Update plan when taking diversions down the road (have to prompt in chat for now for this)
English
Abhishek Malik retweetledi
Deep learning (DL) is proving remarkably effective at solving dynamic models in economics:
📄 nber.org/papers/w33117
Curiously, DL often yields stable solutions without explicitly imposing boundary conditions (e.g., transversality).
Why does this work? A short thread 🧵👇
1/7
English
Abhishek Malik retweetledi
FREE. DEEP. SEARCH.
No bullet points, no lengthy report, no ads, no login, no nonsense design. You ask, we answer. Pure & Deep.
search.jina.ai
English
Carol Dweck’s study shows power of praise:
- Praise talent: Kids avoid challenges to protect their ‘smart’ label
- Praise effort: Kids embrace challenges, work harder, and improve
Effort-based praise builds resilience & a growth mindset. Failures become steps, not setbacks.
Kpaxs@Kpaxs
Want your kids to succeed? Praise effort and grit not talent. Fascinating study.
English
Why it’s important to have contextual usefulness in LLM answer evals…
So we don’t get Captain Obvious answers 😂
Yuchen Jin@Yuchenj_UW
Thank you Gemini, very helpful
English
The future belongs not to those who automate best, but to those who bring the most distinctive vision to life.
Exploring this shift from efficiency to taste in my latest piece: open.substack.com/pub/abhishekma…
English
In 2015, I needed a team to build basic fintech features.
Today, AI lets small teams build entire products.
But here's the plot twist: When everyone has superpowers, your unique perspective becomes your greatest asset.
English
@simonw A micro-service API with the core business logic run partly by an LLM.
English
Let’s see if we can crowdsource a robust definition of “agent” (with respect to AI and LLMs) that fits in a <=280 character tweet
Reply to this with your best attempt, then scroll through the replies and fave the ones that makes sense to you
English
@cursor_ai Have been already using it for a while. Works great!
Feedback: Composer/chat often doesn’t follow import conventions and uses relative path. But this has almost always used conventions. Would love if the other also did the same
English
Abhishek Malik retweetledi
AI voice agents are on 🔥
We're moving from the innovator -> early adopter part of the curve, with new startups sprouting up weekly to serve different verticals.
What @illscience and I are seeing @a16z, and why we're excited 👇
English
AI in sales: What if you could be both? 🤔
Chandler Reed@chandlerreed
There’s only two types of sales guys…
English
LLMs are like gold mines—but we’re barely scratching the surface. The real challenge? Unleashing creativity to build products that don’t just follow the status quo but redefine it.
Love what @cursor_ai folks are doing. There a lot more potential in all workflows!
Sully@SullyOmarr
llms are still criminally under utilized right now there's an element of creativity when it comes to building ai products and i've yet to see anyone crack this for most products, the magic is just the model itself we're in this weird phase where the models are just so damn good they're doing all the heavy lifting, but the products aren't really pushing boundaries
English
Parsing PDFs is still the wild west: unpredictable layouts can wreak havoc on your outputs. Even with the latest AI models, traditional ML and data science preprocessing remains crucial. The ‘bleeding edge’ needs more than just llamas to tame this chaos. 🦙🔍
Jo Kristian Bergum@jobergum
The struggle is real. ColPali is a refreshing alternative imho. From news.ycombinator.com/item?id=410726…
English
Are AI Designers the missing piece to truly harness LLMs’ magic? 🧩 Nick’s point sparks an important discussion on how design and engineering must evolve together to turn these tools into transformative products.
Simon Willison@simonw
“LLMs feel like genuine magic. Yet, somehow we haven’t been able to use this amazing new wand to churn out amazing new products. This is puzzling.” Nick asks if the solution involves a new discipline - AI Designers The design complement to @aiDotEngineer?
English
4/4 This major trend is set to revolutionize AI in 2024.
🔗 Read the full articlto learn more about this exciting shift
bair.berkeley.edu/blog/2024/02/1…
English
2/4 Enhanced control: These systems allow for tighter regulation of AI behavior, improving reliability and trust.
3/4 Tailored performance: Compound systems can be designed to meet specific goals, balancing quality and cost effectively.
English
Berkley AI Research: The Shift from Models to Compound AI Systems
1/4 Compound AI systems integrate multiple tools, enabling real-time data integration and dynamic responses, for top performance and adaptability.
Witnessing this firsthand as we build @EasySLR
English