Mark Hoffmann

Staff ML Engineer @Meta LLMs prev. Ads Core ML | Previously - Chief Architect at Ubiety, ML @NASAJPL @DARPA - neural networks and systems

[Satya Nadella @satyanadella]

https://t.co/vLmiBKTtX3

[Alexandr Wang @alexandr_wang]

1/ today we're releasing muse spark, the first model from MSL. nine months ago we rebuilt our ai stack from scratch. new infrastructure, new architecture, new data pipelines. muse spark is the result of that work, and now it powers meta ai. 🧵

[flavio @flaviocopes]

How Axios was compromised 🤯

[Greg Brockman @gdb]

How AI empowered Paul Conyngham to create a custom mRNA vaccine to cure his dog’s cancer when she had only months to live. The first personalized cancer vaccine designed for a dog:

[Tanishq Kumar @tanishqkumar07]

I've been working on a new LLM inference algorithm.

It's called Speculative Speculative Decoding (SSD) and it's up to 2x faster than the strongest inference engines in the world.

Collab w/ @tri_dao @avnermay. Details in thread.

[Jeremy Howard @jeremyphoward]

WTF is going on at Qwen?!?

Some kind of implosion?

This is really sad and worrying. They've been *such* a strong team, and are losing some of their very best researchers.

[Andrej Karpathy @karpathy]

Gradient descent can write code better than you. I'm sorry.

[Matt Shumer @mattshumer_]

https://t.co/ivXRKXJvQg

[Mark Hoffmann @markkhoffmann]

LLMs for coding are amazing, but my favorite feature is actually being able to talk about any subject and instantly be able to dive into the fine grained details of learning.

I just deep dove into some of the best papers of NeurIPs from the past year and the rate of information transfer is night and day difference to even a year ago.

[Mark Hoffmann @markkhoffmann]

@YangjunR @thinkymachines Would love to chat!

[Mark Hoffmann @markkhoffmann]

I’ll be at NeurIPS this week, DM me if you want to meet up!

[Mark Hoffmann @markkhoffmann]

Another honorable mention from a sprint last year: https://t.co/u7wjE9g1sD

[Mark Hoffmann @markkhoffmann]

Does anyone else go full flow state where you are coding something and end up needing to listen to the same song on repeat. But it’s also the first thing you put on when you wake up, in the shower, on the drive to work, while at work, while at the gym, later at night, before going to bed.
Only happens 2-3 times a year, but it lasts for about two weeks at a time and you listen to the same song >1k times.

[Mark Hoffmann @markkhoffmann]

Current song: https://t.co/m2L8h1DEJm

[Mark Hoffmann @markkhoffmann]

@Beyond_AK @ItzSuds Agree, surprisingly a lot of stuff to do. I ended up in Redwood Shores right off of a ton of really scenic trails.

[Mark Hoffmann @markkhoffmann]

Agree at smaller and at most companies it's overkill. When you are operating at a certain scale and defendable metric driven impact is in the company's culture, it's necessary and table stakes.
"revenue went up" doesn't cut it when you have to justify budget allocation for capacity being split among dozens of teams.

[Mark Hoffmann @markkhoffmann]

Metric conversions are hard… let’s convert quality improvement to OpEx they said.. create a whole framework around precision and recall improvements translating then realize you have to do an even more involved derivation for free form rubric graded output of measuring probability of samples batch pass rates via binomial distribution priors…

At least I felt like I was solving the gravity equation from interstellar for awhile on a large whiteboard

[Mark Hoffmann @markkhoffmann]

@ClickHouseDB finally hunted this down to a compression incompatibility issue. Athena requires snappy compression for parquet files vs. clickhouse default export which is LZ4

[Mark Hoffmann @markkhoffmann]

@ClickHouseDB can you fix your export parquet to s3 functionality? It shouldn’t be this hard to export files and then query them with Athena. Whatever is happening under the hood causes HIVE_CURSOR_ERROR issues.

Doesn’t happen when writing parquet files to s3 with dask though

[Mark Hoffmann @markkhoffmann]

@dmartinezg05 Ah I usually go with chicken which seems pretty full still :)

[Mark Hoffmann @markkhoffmann]

How has all food doubled in price since 2021 except Chipotle? What kind of magic do they have to keep their supply chain as the only one untouched?