in search of the global minima. always hill climbing, always truth seeking.
Singapore
- Tweets 2,061
- Following 3,248
- Followers 516
- Likes 17,864
There is no AI without data. Have a deep understanding in data engineering pays dividend and compounds along the way.
Data engineers > AI engineers
there's a big difference between a "RAG guy" and a "Data Intensive Applications" person like in this screenshot and there's never been a starker difference in demand in my entire career in tech (aka post search engine wars)
if your blog / github is full of stuff like this, I have a fantastic mission/job for you (SF preferred, can relocate u, visas ok)
(if you can name who's blog this is without even checking, you definitely are in the community. bonus points if you know where he ended up also without looking)
jobs@latent.space and the team will reach out if you fit
1
28
Moving Terminal-Bench 4.0 from 31.2 to 42.3 while keeping the same version number and calling it "minor update" is incredible.
Makes you wonder what an actual minor and major version bump could bring!
🚀 DeepSeek-V4.1-Flash-1017 is here!
Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our website, app, and API.
🛠️ Optimized for @DeepSeekHarness, Tencent CodeBuddy, WorkBuddy, and more. Upgrade to DeepSeek Harness v0.2.1-rc.1 or later for the best experience with the new model.
1/4
4
73
Derek Chia retweeted
50,000 stars on GitHub ⭐
Thanks to everyone who’s built with ClickHouse, contributed, or helped someone get started.
Curious what the fuss is about? Try ClickHouse Cloud and run your first query → buff.ly/VsnWZ0j
1
7
23
1,351
Build your own harness, folks.
This is absolute banger paper from NVIDIA on self-evolving agent harnesses.
(bookmark it)
They introduce SoL-Pi which cuts token traffic by nearly half.
And it matches its baseline harness on GPT-5.6 Sol and Opus 5.
More details below:
Instead of tuning a harness by hand, they run auto-research loops at the harness layer across many repository-derived and verifier-driven environments, keeping only the mechanisms that survive selection.
Four mechanisms survived:
> Action Fusion changes how actions execute
> Online Context Compact handles compaction during a run
> ObservationPack reshapes observation handling
> Evidence-Preserving Reducer covers delegated reading
On the 51-task EdgeBench evaluation, the savings translate to about a third off API cost. In dollars that is an estimated $8.75 to $13.50 per hour against native Codex and Claude Code harnesses, and $4.36 to $5.71 against the baseline harness.
Because the search runs across many environments rather than one, the retained mechanisms keep working outside the setting that produced them. Code is on GitHub under NVlabs.
Paper: arxiv.org/abs/2609.20519
Chat with Paper: academy.dair.ai/papers/sol-p…
113
359
3,260
171,792
Jev is essentially a super good (and fast) zero-shot classifier. This reminds me of the good old days of BERT/BART and spending countless hours on huggingface - huggingface.co/tasks/zero-sh…
1
101
Derek Chia retweeted
Singapore, our next meetup is headlined by Max Deichmann, Co-founder & CTO of Langfuse 🎤
Building AI Products
🗓️ Friday, Sept 25 · 6:00 PM · AWS Office | Singapore
Langfuse is the open-source LLM engineering platform used by thousands of teams to trace, evaluate, and debug AI applications — and it runs on ClickHouse under the hood.
Max's talk, "Trace It, Store It, Ship It," goes past the demo stage into what production actually demands.
Seats are limited and registration requires approval: buff.ly/zLpT5ZQ
Want to speak at a future APJ meetup? buff.ly/ak4tJO8
#ClickHouse #Langfuse #Singapore #LLM #AIEngineering #Observability
2
7
1,286
The Foreign Minister of Singapore just shared his Agent-Maintained Knowledge Graph. I guess we all need to catch up!
Agent-Maintained Knowledge Graph for Context Retrieval using OKF and SQLite has improved my agents’ query salience and saved tokens (not re-reading half the wiki to answer a question)
gist.github.com/VivianBalakr…
1
2
130
A terminal multiplexer is what we need in this world with abundance intelligence.
I've started a new company: @superlogical! We're going to begin by building a terminal multiplexer. The entire vision is much larger, but the multiplexer is the foundation. Sign up for the newsletter to get beta access and devlogs (product updates only I promise).
superlogical.com/
1
99
Derek Chia retweeted
ClickHouse x @FulhamFC ⚽
We're proud to be the Club's new principal partner and front-of-shirt sponsor. You'll see us on the home, away, and third kits this season, and at Craven Cottage all year.
We look forward to welcoming our customers, partners, and community to the Cottage this season!
5
17
125
40,450
Heads up to all tech and data geeks in Singapore! Join us at Open House roadshow to learn more about the real-time data platform for AI! 👇
ClickHouse is bringing its Open House roadshow to Singapore on Sept. 24!
Join Krithika Balagurunathan (ClickHouse), Max Deichmann, CTO of Langfuse, and other global experts for practical strategies and real-world AI use cases.
1
57
Boris, head of Claude Code, shares his latest thinking on where Claude Code is headed — read on to see why his take matters. piped.video/watch?v=Z47vatps…
AI coding isn’t the story anymore — the real story is that the bottleneck has moved.
The big shift isn’t code generation anymore. It’s deciding what matters, reviewing it safely, shipping it, and turning it into business value.
2
2
111
Practical tip: use exploratory output
When you want better reasoning, more options, and clearer thinking, use `/config output-style exploratory`. It helps the model surface possibilities instead of jumping straight to a final answer. That makes it especially useful for planning, debugging, and idea generation.
1
1
11