Maria Gorinova's Avatar

Maria Gorinova

@mgorinova.bsky.social

Shaping the future of programming @tessl.io πŸš€ | ex-@TwitterCortex @Birdwatch πŸ’™ | PhD in probabilistic machine learning, loyal servant to a cat, collector of random variables, and lover of well-placed puns. https://mgorinova.github.io/

2,123 Followers  |  433 Following  |  131 Posts  |  Joined: 01.07.2023  |  2.5849

Latest posts by mgorinova.bsky.social on Bluesky

Preview
a raccoon laying on a bed with a bowl of blueberries in its mouth Alt: a raccoon laying on a bed with a bowl of blueberries in its mouth; labelled "bluwbewwy"

"I really need a tool that helps me count the number of letters in 'blueberry'. It will change my life."

... said no one ever.

10.08.2025 08:38 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0
Preview
a pirate hat with a skull and crossbones and a sword on it Alt: a cat with a pirate hat with a skull and crossbones and a sword on it
03.08.2025 14:22 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

I stole this for the linkedin post πŸ˜‚ Gratitude 🫑

03.08.2025 13:42 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Aaaaaaa! I'm regretting my choices πŸ₯²

03.08.2025 09:06 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

I actually like the em dash. Am I an LLM? πŸ€”

03.08.2025 08:55 β€” πŸ‘ 18    πŸ” 0    πŸ’¬ 2    πŸ“Œ 2
A chatgpt generated take on the iconic kiss on the Berlin wall. But with Elon Musk kissing the Twitter blue bird. The text reads "My God, help me survive this tweet love"

A chatgpt generated take on the iconic kiss on the Berlin wall. But with Elon Musk kissing the Twitter blue bird. The text reads "My God, help me survive this tweet love"

A few years late but I was finally able to generate this

26.05.2025 17:46 β€” πŸ‘ 6    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0
Preview
a man is holding a piece of bread over a woman 's face and asking what are you ? ALT: a man is holding a piece of bread over a woman 's face and asking what are you ?

What I imagine open ai feels like after reading that s1 paper

11.02.2025 00:17 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Wait... So s1 cost only $50 to fine-tune and beat o1-preview. And the secret sauce is... forcing the model to generate "Wait" instead of end-of-sequence???

Ahahhahahhaha this is so cool

11.02.2025 00:16 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

🚨 MAJOR ALERT 🚨

The people wanted verification badges on Bluesky and @guan.dk and I have teamed up to bring you verification badges.

Behold! It is the Official Verified Labeller!
bsky.app/profile/veri...

26.11.2024 04:24 β€” πŸ‘ 2809    πŸ” 643    πŸ’¬ 255    πŸ“Œ 153

Strong characters sounds amazing + I think there is a lot to learn from that time if portrayed well. Thanks for the recommendation, I will give it a go!

25.11.2024 00:54 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Hi John,

That’s what all the bots say πŸ˜³πŸ˜‚

Not really my area of expertise, I know of the very obvious ones like input sanitation, regular finetuning on problematic examples, etc. The folks at lakera.ai have been on it for some time and regularly host hacking challenges, which I enjoy following

25.11.2024 00:51 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 2    πŸ“Œ 0

I love Silo and For All Mankind. Maybe I should try Halt & Catch Fire next!

25.11.2024 00:40 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Thank you! Indeed, what a debacle πŸ˜‚ 🍿

24.11.2024 00:20 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0
Post image

I think this chart better conveys the benefit and costs of a specialized model

23.11.2024 17:33 β€” πŸ‘ 54    πŸ” 12    πŸ’¬ 5    πŸ“Œ 1

What's the drama with Byzantine? Did someone raise it during reviewing in the past? Is there a public discussion? 🍿

23.11.2024 07:10 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Bluesky's firehose is a treasure trove of public data for researchers and developers, and it's completely free. Check out our developer docs: docs.bsky.app

23.11.2024 05:54 β€” πŸ‘ 7982    πŸ” 1548    πŸ’¬ 324    πŸ“Œ 171
Book outline

Book outline

Over the past decade, embeddings β€” numerical representations of
machine learning features used as input to deep learning models β€” have
become a foundational data structure in industrial machine learning
systems. TF-IDF, PCA, and one-hot encoding have always been key tools
in machine learning systems as ways to compress and make sense of
large amounts of textual data. However, traditional approaches were
limited in the amount of context they could reason about with increasing
amounts of data. As the volume, velocity, and variety of data captured
by modern applications has exploded, creating approaches specifically
tailored to scale has become increasingly important.
Google’s Word2Vec paper made an important step in moving from
simple statistical representations to semantic meaning of words. The
subsequent rise of the Transformer architecture and transfer learning, as
well as the latest surge in generative methods has enabled the growth
of embeddings as a foundational machine learning data structure. This
survey paper aims to provide a deep dive into what embeddings are,
their history, and usage patterns in industry.

Over the past decade, embeddings β€” numerical representations of machine learning features used as input to deep learning models β€” have become a foundational data structure in industrial machine learning systems. TF-IDF, PCA, and one-hot encoding have always been key tools in machine learning systems as ways to compress and make sense of large amounts of textual data. However, traditional approaches were limited in the amount of context they could reason about with increasing amounts of data. As the volume, velocity, and variety of data captured by modern applications has exploded, creating approaches specifically tailored to scale has become increasingly important. Google’s Word2Vec paper made an important step in moving from simple statistical representations to semantic meaning of words. The subsequent rise of the Transformer architecture and transfer learning, as well as the latest surge in generative methods has enabled the growth of embeddings as a foundational machine learning data structure. This survey paper aims to provide a deep dive into what embeddings are, their history, and usage patterns in industry.

Cover image

Cover image

Just realized BlueSky allows sharing valuable stuff cause it doesn't punish links. 🀩

Let's start with "What are embeddings" by @vickiboykis.com

The book is a great summary of embeddings, from history to modern approaches.

The best part: it's free.

Link: vickiboykis.com/what_are_emb...

22.11.2024 11:13 β€” πŸ‘ 654    πŸ” 102    πŸ’¬ 22    πŸ“Œ 6

πŸ’― no fake science please 😬😬😬 This sort of accounts can also be used to build up a following base and then activate as part of an influence campaign

22.11.2024 09:09 β€” πŸ‘ 4    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Personally, I like Quiet Posters: it surfaces posts I wouldn't have seen otherwise!

22.11.2024 01:32 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

By the authority vested in me by the decentralised gods (πŸ’…πŸ») I accept this proposition.

(πŸ’…πŸ» Sarcasm. I possess no such authority. The decentralised gods would never grant it to me (πŸ’…πŸ»))

20.11.2024 09:09 β€” πŸ‘ 1    πŸ” 1    πŸ’¬ 1    πŸ“Œ 0

Maybe we need a special sarcasm tag here on bsky

20.11.2024 08:51 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

My question was a joke!

I know what the word means 🀣

20.11.2024 08:50 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 2    πŸ“Œ 0

What do you mean "once"? 🀨

20.11.2024 08:32 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

I got a Probabilistic Programming starter pack going. Hit me up if you're involved with #probprog R&D and want in!

go.bsky.app/JfvubEf

19.11.2024 16:10 β€” πŸ‘ 116    πŸ” 30    πŸ’¬ 23    πŸ“Œ 1

Thanks, Chad! Much needed starter pack πŸ™ŒπŸ»

19.11.2024 19:37 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Bluesky tip for newbies: Bluesky has no real algorithm. Likes here do nothing but show the poster that you enjoyed their content.

That kindness is always going to be appreciated by creators, but if you want to help to boost something here, you have to repost it so others see it. πŸ™‚

18.11.2024 10:27 β€” πŸ‘ 26518    πŸ” 9220    πŸ’¬ 20    πŸ“Œ 12

Yet another plea to people who are using or following people off follow-farming hashtags like #NoDemUnder1K, #NoDemUnder5K, #BlueCrew, #FBR, etc etc: please do not do that, it's how disinformation operations work. I'm already seeing accounts in those hashtags that are likely disinformation ops. 1/

15.11.2024 14:28 β€” πŸ‘ 4134    πŸ” 2001    πŸ’¬ 173    πŸ“Œ 343

Hey kids we brought @pivotpod.bsky.social over too.

bsky.app/profile/pivo...

18.11.2024 20:45 β€” πŸ‘ 1052    πŸ” 89    πŸ’¬ 84    πŸ“Œ 5
Video thumbnail

A growth rate of ~10 users per second 🀯

Great dashboard by @natalie.sh!

bcounter.nat.vg

18.11.2024 13:53 β€” πŸ‘ 31    πŸ” 3    πŸ’¬ 1    πŸ“Œ 1

@mgorinova is following 20 prominent accounts