λndres Mariscal @SerialDev

Wrote anti-cheat ml, do ML/AI at places you know off and probably use; systems ML person who still cares about cycles, graphics, compilers in the agentic era Helsinki Joined July 2015

Tweets

3K
Followers

333
Following

3K
Likes

15K

λndres Mariscal @SerialDev

a month ago

. With agentic coding, I think this approach will be a lot more common! gamedev used to do it quite a lot like it too, I remember we had everything from re-implementations of stdlib stuff too for our anticheat. TigerBeetle really resonates with good engineering imo

Joran Dirk Greef @jorandirkgreef

a month ago

TigerBeetle has a strict zero dependencies rule (apart from Linux and the compiler).

35 22 489 25K 45

0 0 1 65 1

View Details

λndres Mariscal @SerialDev

a month ago

@lemire I would think so, but we are on a crunch cycle in the industry, when hiring picks back up it will be a disaster

0 0 0 3 0

View Details

λndres Mariscal @SerialDev

a month ago

@rfleury Non-AI content to read 🫡 honestly miss that a lot lately

1 0 4 212 0

View Details

λndres Mariscal @SerialDev

2 months ago

@halvarflake hard to control for post-covid, many things are demonstrably worse. Sadly it leaks to our personal life sometimes

0 0 1 565 0

View Details

λndres Mariscal @SerialDev

2 months ago

@c00ldata @rfleury @TomWatkins1994 I love zig, but Jai's compilation speed alone is the absolute best feature imo

0 0 0 77 0

View Details

λndres Mariscal @SerialDev

3 months ago

@kellabyte I'm a huge fan of how transparent they are, I've learned a lot through their talks and interactions here. Truly a great team

0 0 3 966 0

View Details

λndres Mariscal @SerialDev

3 months ago

I didn't know you started tirso with zig! I'm a big fan of niche languages, jai is incredible for performance focused stuff , but in an LLM driven world data will have such a strong impact long term. However I do believe niche languages will shine through given they hold a higher distribution of high performance well written code. Agree with your points just hope the hobbyist languages don't fade out completely

0 0 1 387 0

View Details

λndres Mariscal @SerialDev

3 months ago

Its an inefficiency on both ends tbh, often you are also met with people that cannot understand the nuance of what you built or the circumstances that led to certain decisions, or even understand it at a reasonable level. I think applicants often have less context on employers too.

0 0 1 81 0

View Details

λndres Mariscal @SerialDev

3 months ago

@mcnabbd That's my preference tbh, if you are hiring people good at something let them, and more often than not the outcome will be far better than if you managed every interaction. I bet it was a success

1 0 2 48 0

View Details

λndres Mariscal @SerialDev

3 months ago

Getting funding for anything non-trivial is just not feasible in Europe most of the time, one of my favourite personal anecdotes, we were looking for funding on a young startup I was hire 4 in; 10 years ago, no European VC would take us even though we were cashflow positive at that stage and growing. we got acquired months later

0 0 5 482 1

View Details

Joran Dirk Greef @jorandirkgreef

3 months ago

At TigerBeetle, as systems engineers, we: 1. write code, sharpening skill and second order understanding 2. read code, walking through the building to constantly improve things 3. review code, ensuring no code is merged without two engineers’ understanding and review.

18 42 437 17K 83

View Details

λndres Mariscal @SerialDev

3 months ago

@jorandirkgreef It is still wild to me this is something we have to state now! I lowered my guard spotted some rust and back at working this way too

0 0 1 333 0

View Details

λndres Mariscal @SerialDev

3 months ago

@halvarflake have you had a chance of testing it with Fable? I also find Opus not that great on verifiable non-iterative work

0 0 0 66 0

View Details

ハセン حسن @hasen_95dx

3 months ago

Managers will literally fire their best engineers if they fear they threaten their position. It's all politics. Maximizing shareholder value my ass.

Justin Poehnelt @JPoehnelt

3 months ago

Two months ago I was fired by Google for creating the Google Workspace CLI. It went viral, hit #1 on Hacker News, gained thousands of GitHub stars and many thousands of actual users in just a couple days. It was an incredible, confusing journey, from directors and leaders asking

607 1K 15K 4.9M 5K

0 2 38 1K 4

View Details

λndres Mariscal @SerialDev

3 months ago

@jeremyphoward . They did, there is different baselines to most AI companies at the moment on a philosophical level, ex-effective altruism groups, accelerationism, etc. if you are familiar with the sub-cultures which I know you are, you know very well that this was well orchestrated.

0 0 1 394 0

View Details

Daniel Lemire @lemire

4 months ago

Nvidia is proposing a beast of a CPU system for Windows PCs. It has 128 GB of shared memory and comes with up to 6,144 state-of-the-art CUDA cores. CPU wise, the chip has 10 performance cores and 10 efficiency cores. The performance cores are based on the Cortex-X925. These chips appear to support six 128-bit SIMD execution units (SVE2), not as good as recent AMD chips, but better than Apple Silicon (on paper). The game changer is the unified 128 GB memory. That is the path Apple took years ago. Instead of separate memory for the CPU and GPU, everything shares a single pool. It is increasingly popular. The memory is not as fast as dedicated GPU memory, but it is cheap enough while delivering enough bandwidth to run AI models locally. I am not sure how many people will run AI models locally. It still seems like a niche application to me. However, it will make decent machines to play video games. It will be interesting to see how Intel and AMD respond. I think that the AVX-512 instructions supported by all recent AMD processors are far superior to the SVE2 instructions of the Cortex-X925. They can eat more data and they are more versatile. But Intel has been shy, thus far, in making it available on customer systems.

37 21 191 24K 58

View Details

λndres Mariscal @SerialDev

4 months ago

@Jonathan_Blow Arm is involved on the tweet sequence too, so I would expect unified memory and built-in tpu or something AI-centric

1 0 0 1K 0

View Details

λndres Mariscal @SerialDev

4 months ago

@stake_jevens @luminal_ai Location agnostic?

0 0 0 137 0

View Details

λndres Mariscal @SerialDev

4 months ago

@lucasmeijer Token attention capacity is still limited, increased context window is not truly increased imo, just look at how its implemented, Grouped-Query Attention ;; RoPE Scaling ;;YARN / CoPhy, theoretically do compress information so it is in essence lossy. less is more