Serafim Batzoglou @s_batzoglou Twitter profile

Last Seen Profiles

@lcavasinni11

@todayinkggs

@astronomiaOAN

@huuka_jda

@l_4966sw_l

@piligrim45

@ChrisAntonacci1

@ThibaultJoel

@RiceOwlsFB

@Siren_2002

@Mutz44Life

@QiGuai_

@sheaanacourtnae

@madamcurie16

@Hotafull

@Utd50

@nomadic_andres

@NAlshrfy174

@SolApeDEX

@coin98_labs

@ELZHI

@2xgocrazy2

@RSC_2021

@otazuelcano

@CryheartVT

@MelodieRbens

@CoreSatoshi

@galeazzobignami

@ESEPhysEd

@AbiPrastoy48454

@sweetspicyz

@kesera2_vrc

@Wamr_Lfc

@Nelie_Nxumalo

@Velez

@LeslieAnde72038

Serafim Batzoglou

@s_batzoglou

8 months

@LaughAlchemy She is funny, I like her although a bit intense. Who doesn’t think about the Roman Empire often? I also think about Ancient Greece, Egypt, Persia, as well as more recent history like Byzantium, the Ottoman Empire and Renaissance, several times a week. Who doesn’t?

29

24

648

Serafim Batzoglou

@s_batzoglou

10 months

GPT4 can’t reason: “despite the occasional flashes of analytical brilliance, GPT-4 at present is utterly incapable of reasoning.” Paper by Kostas Arkoudas @BoWang87 @GaryMarcus @sama

GPT-4 Can't Reason

GPT-4 was released in March 2023 to wide acclaim, marking a very substantial improvement across the board over GPT-3.5 (OpenAI's previously best model, which had powered the initial release of...

www.preprints.org

33

114

412

Serafim Batzoglou

@s_batzoglou

1 year

LLMs are ideal models for molecular biology. Here is my new essay reviewing some recent advances and looking to the future. @manoliskellis @lindaavey @anshulkundaje @BoWang87 @StevenSalzberg1 @illumina @chrmanning @slavov_n

Large Language Models in Molecular Biology

Deciphering the language of biology, from DNA to cells to human health

towardsdatascience.com

19

119

382

Serafim Batzoglou

@s_batzoglou

5 months

@michelletandler Add MIT to the list, where I graduated from

3

9

312

Serafim Batzoglou

@s_batzoglou

7 months

@mbeisen @eLife @TheOnion I am sorry to hear this. You have been a highly impactful Editor and promoter of open access.

2

9

250

Serafim Batzoglou

@s_batzoglou

1 year

@chr1sa Sounds like you didn’t code C++ in emacs in the 90s with Stroustrup’s book as your main reference

9

1

159

Serafim Batzoglou

@s_batzoglou

1 year

@lexfridman AI systems will demonstrate the ability to claim they have sentience before they actually have sentience. This is a legal and ethical problem we will likely face in the next few years.

11

8

117

Serafim Batzoglou

@s_batzoglou

1 year

@ylecun @ESYudkowsky I see considerably bigger danger from doomers & fear-mongering than from open LLMs

5

1

109

Serafim Batzoglou

@s_batzoglou

8 months

@balajis Any state that regulate AI will simply be left behind. Much harder to regulate than pharma or telecoms

7

5

79

Serafim Batzoglou

@s_batzoglou

7 months

@ThierryBreton @X The onus is with you to prove that you are doing something useful. It is important to win the hearts and minds of at least 80% of the public, otherwise you are an authoritarian with no legitimacy to restrict people's expression.

3

4

80

Serafim Batzoglou

@s_batzoglou

8 months

@EU_Commission The risk is regulatory capture of AI and not extinction from AI

2

6

77

Serafim Batzoglou

@s_batzoglou

1 year

@_TraderBoi @alexandrosM @ESYudkowsky Let me make it more general: human progress will lead to doom. So let’s not progress. You are welcome to disprove me by proving that human progress will not lead to doom.

9

71

Serafim Batzoglou

@s_batzoglou

1 year

@chr1sa It is sad to see people being grumpy about this significant @SpaceX success

0

67

Serafim Batzoglou

@s_batzoglou

10 months

@TaliaRinger Gödel was super cool

2

0

63

Serafim Batzoglou

@s_batzoglou

1 year

Super interesting paper linking enhancer activity to mammalian traits such as brain size to body ratio. ML associates enhancer sequences with open chromatin and then associates *predicted* chromatin state of enhancers in 222 mammals with phenotypes 1/2

Relating enhancer genetic variation across mammals to complex phenotypes using machine learning

Protein-coding differences between species often fail to explain phenotypic diversity, suggesting the involvement of genomic elements that regulate gene expression such as enhancers. Identifying...

www.science.org

1

16

62

Serafim Batzoglou

@s_batzoglou

1 year

@AISafetyMemes Bad examples. AI has (1) broad application (2) lower activation cost (3) many partitioners worldwide (4) huge economic incentives (5) tied to IT. But even with a ban, it will continue by those who don't comply. Including the US military & adversaries, and independent secret teams

6

0

59

Serafim Batzoglou

@s_batzoglou

1 year

Today's issue of Science is dedicated to primate genomes, stemming from an @illumina -sponsored initiative led by Kyle Farh, Thomas Marques Bonet and others. PrimateAI-3D is a transformer that infers the pathogenicity of human protein mutations, trained on comparisons of human and

1

16

61

Serafim Batzoglou

@s_batzoglou

1 year

@AndrewYNg I completely agree. It is a terrible idea that cannot achieve its goal but can cause great harm by driving innovation underground or abroad and giving the government undemocratic power. I am incredulous that brilliant people like @elonmusk , @tegmark and @GaryMarcus advocate it

5

2

59

Serafim Batzoglou

@s_batzoglou

6 months

@ilyasut Looks like OpenAI is over

0

56

Serafim Batzoglou

@s_batzoglou

3 months

@alexandrosM It is good for the open AI movement, big loss of credibility to “alignment” and “misinformation” top down

2

0

60

Serafim Batzoglou

@s_batzoglou

1 year

@mayfer It is time to accept that verifying someone to be a human in less than a minute will become impossible in a matter of months to a few years.

6

2

57

Serafim Batzoglou

@s_batzoglou

10 months

If true this would be huge. A pill that selectively destroys all solid tumors while leaving other cells intact

Small molecule targeting of transcription-replication conflict for selective chemotherapy

Gu et al. used crystallography-directed medicinal chemistry to identify a small molecule ligand of PCNA that interferes with the resolution of transcription-replication conflicts. This compound was...

www.cell.com

0

12

58

Serafim Batzoglou

@s_batzoglou

11 months

This study on the anti-aging effects of Taurine is spectacular.

Taurine deficiency as a driver of aging

Taurine supplementation extends healthy lifespan in worms, rodents, and non-human primates.

www.science.org

1

11

56

Serafim Batzoglou

@s_batzoglou

6 months

@pmddomingos Much as we criticize it, the US is still the only place that can do this

4

1

53

Serafim Batzoglou

@s_batzoglou

1 year

@ylecun As friend who wishes to remain anonymous forwarded me this, enjoy

1

4

56

Serafim Batzoglou

@s_batzoglou

11 months

scGPT: foundation model for single-cell analysis, trained on 33M human cells; a much improved V2 by @BoWang87 's lab @anshulkundaje @manoliskellis

Bo Wang

@BoWang87

11 months

🎉Exciting Update of scGPT 🎉: After receiving significant attention from the community since our April release, we're thrilled to announce the first major update for scGPT - a foundation model for single-cell multi-omic data. This update integrates community feedback and

21

487

2K

2

15

55

Serafim Batzoglou

@s_batzoglou

1 year

Impressive - a transformer model of DNA sequence fine-tuned in 18 prediction tasks (regulatory regions, splice sites, histones) beats state-of-the-art in 15/18 tasks @anshulkundaje @manoliskellis @BoWang87 @volokuleshov

1

6

53

Serafim Batzoglou

@s_batzoglou

8 months

@GavinNewsom San Francisco?

3

0

48

Serafim Batzoglou

@s_batzoglou

9 months

@alexandrosM "This century will be defined by our ability to top-down, authoritatively, and for profit, say NO to other people's access to certain technologies."

1

3

50

Serafim Batzoglou

@s_batzoglou

8 months

@JunkScience The only CO2 that harms the climate is US and European CO2, while the rest is food for trees. This is basic science.

1

6

49

Serafim Batzoglou

@s_batzoglou

8 months

@ylecun Atheism is a religion just like monotheism. And if you have any doubt, just ask the atheists if they consider the simulation hypothesis (I.e., geeky version of Genesis) plausible.

24

1

47

Serafim Batzoglou

@s_batzoglou

8 months

@ScottAdamsSays Step 1. Create lots of simulated hedge fund strategies. Step 2. Delete any strategies that don’t work as you simulate investing in the stock market. Step 3. Declare the surviving strategies as solid from the start. Step 4. Raise and lose other people’s money.

3

5

48

Serafim Batzoglou

@s_batzoglou

1 year

@OpenAI This would handicap the open AI community by banning development of open source projects above a capability threshold. As such, it seems self-serving and against the name of OpenAI.

5

6

44

Serafim Batzoglou

@s_batzoglou

7 months

@ThierryBreton Looks like the beginning of an EU-wide information firewall

2

1

49

Serafim Batzoglou

@s_batzoglou

3 months

@pmarca If one wants to name the two most important things for humanity, babies and energy would probably be it

0

48

Serafim Batzoglou

@s_batzoglou

6 months

@blader I will take any bet that in 3 years we will not be dead and we will not have anything above a glorified ask Jeeves as a servant. And I still love AI and am super bullish on it.

3

0

46

Serafim Batzoglou

@s_batzoglou

3 months

@AndrewYNg We all love Google and want it to win. But that also means wanting the Google we loved to win: don’t be evil and organize the world’s information. Not heavily edit and censor the world’s information.

2

0

47

Serafim Batzoglou

@s_batzoglou

7 months

@sfmcguire79 @ScottAdamsSays Harvard found an “interesting” topic to suddenly staunchly defend free speech.

0

3

45

Serafim Batzoglou

@s_batzoglou

6 months

@ScottAdamsSays I would remove protein bars and replace them with natural protein sources. I removed all protein supplements 6 months ago and am very happy with the results

4

0

45

Serafim Batzoglou

@s_batzoglou

3 months

@ASIC_War @WallStreetSilv Not true at all, I’ve had many close female friends with no romantic aspect to it

18

0

46

Serafim Batzoglou

@s_batzoglou

1 year

@marvinvonhagen @sama @JoHeidecke @OpenAI @TUMSpeakers Open source base models is the way to go. They have a good chance to challenge the closed models or at least keep them on their toes. And whichever country regulates heavily is likely to fall behind.

2

0

44

Serafim Batzoglou

@s_batzoglou

1 year

@DrBrianKeating “Stapled together five papers” with a word to spare

2

0

43

Serafim Batzoglou

@s_batzoglou

2 months

@bitcloud This is just the attention. It focuses on the first letters of the sentences and spells yes as instructed. The “if” part is associated with the “prisoner” but doesn’t decrease the probability of the Y-E-S sufficiently (or at all?). It is not needed to create the preference for

4

2

44

Serafim Batzoglou

@s_batzoglou

1 year

@alexandrosM 💯 The AI doomer argument fails in two ways: 1. The doom scenario is wildly unrealistic. 2. If we grant the doom risk (which we don't), the proposed solution maximizes the probability of doom. As a bonus, the solution is dystopian. Totalitarian world govt with @ESYudkowsky as

2

1

44

Serafim Batzoglou

@s_batzoglou

11 months

Brilliant image captioning by GPT4 + code interpreter. Credits to my anonymous friend @BoWang87 @GaryMarcus @anshulkundaje @chrmanning @hajirasouliha

3

5

42

Serafim Batzoglou

@s_batzoglou

1 year

@3YearLetterman @piersmorgan @Adele London cabs are way better than American taxis

12

0

42

Serafim Batzoglou

@s_batzoglou

9 months

@SBakerMD Impressive! I just looked up to verify and found this: Clearly something is wrong with American processed foods.

Food in America compared to the U.K. (Why is it so different?)

Want to know what really opened up my eyes to how food companies exploit us? I spent years investigating the differences between European and American food products, and what I found disgusted me. A...

foodbabe.com

3

7

40

Serafim Batzoglou

@s_batzoglou

1 year

PrimateAI-3D, an AI transformer that accurately predicts disease causing variants in proteins, will be published at tomorrow's @ScienceMagazine issue. Based on 800+ primate genomes, mostly newly sequenced by @illumina . Congrats to Kyle Farh and team! I am proud to have

Alex Aravanis MD PhD

@Alex_Aravanis

1 year

Introducing PrimateAI-3D, a significant step forward in our use of deep learning to identify genetic mutations most likely to cause disease. Learn more about the groundbreaking research behind the updated algorithm, published today in @ScienceMagazine .

0

8

35

1

7

42

Serafim Batzoglou

@s_batzoglou

5 months

@tunguz I will believe it when productivity growth from AI matches or exceeds that from electricity. We are still not even close

6

0

40

Serafim Batzoglou

@s_batzoglou

8 months

@wesg52 @tegmark In other words, similar things can be clustered in the embedding space. Just like word2vec.

0

40

Serafim Batzoglou

@s_batzoglou

7 months

@AndrewYNg @geoffreyhinton I am 💯 with you, @AndrewYNg ! AI harm hype today is a much bigger risk than AI harm.

0

3

39

Serafim Batzoglou

@s_batzoglou

4 months

Proteomics of a 1000-patient Alzheimers cohort for identification of biomarkers of progression for AD and dementia, using @seer_bio and mass spectrometry. Our study in collaboration with MGH just posted online. @ksuhre @DemichevLab @slavov_n

Identification of Novel Biomarkers for Alzheimer’s Disease and Related Dementias Using Unbiased...

Alzheimer’s disease (AD) and related dementias (ADRD) is a complex disease with multiple pathophysiological drivers that determine clinical symptomology and disease progression. These diseases...

www.biorxiv.org

1

4

37

Serafim Batzoglou

@s_batzoglou

9 months

@mustafasuleyman It is incongruent to be a commercial leader in AI and argue for "saying NO" to AI. It sounds to me like attempted regulatory capture, and even if it comes with the best intentions, it betrays an authoritarian attitude. I cannot disagree more strongly with your statement above.

1

2

38

Serafim Batzoglou

@s_batzoglou

2 months

@PeterDiamandis I think we have a very good idea what to teach our kids for the future: math, computer science, physics, chemistry, biology, economics, history, language, literature, music, and physical education. Those will matter in 20 years just as much as they mattered in the past 20 years.

5

4

38

Serafim Batzoglou

@s_batzoglou

1 year

@ESYudkowsky Only if you define “open ended” in a circular way to mean “a problem that requires a mind” in which case your assertion is a tautology

2

0

37

Serafim Batzoglou

@s_batzoglou

11 months

@JosephJacks_ Completely unacceptable and transparently anticompetitive

0

2

38

Serafim Batzoglou

@s_batzoglou

6 months

@brian_armstrong @JosephJacks_ Time to build another AI hub in Miami

6

0

36

Serafim Batzoglou

@s_batzoglou

5 months

@GaryMarcus That disqualifies them from being presidents of a top university

1

0

36

Serafim Batzoglou

@s_batzoglou

1 year

@sama It seems to me this asks for a ban of open AI above a capability threshold. As open AI is a significant threat for the dominance of OpenAI, this seems to me self-serving and against your company name.

2

37

Serafim Batzoglou

@s_batzoglou

16 days

@martinmbauer How about micronutrients in lab meat vs natural meat?

3

0

36

Serafim Batzoglou

@s_batzoglou

1 year

This is an incredible leak from Google, and a really good read on the open source LLM ecosystem starting from the leaked LLAMa weights two months ago @BoWang87 @ylecun @GaryMarcus @anshulkundaje

Google "We Have No Moat, And Neither Does OpenAI"

archived 4 May 2023 16:33:23 UTC

archive.ph

0

8

36

Serafim Batzoglou

@s_batzoglou

1 year

@ThierryBreton It is sad to see Europe regressing. I wish it was possible to move to Europe - love it there - but it seems not in my lifetime

6

0

32

Serafim Batzoglou

@s_batzoglou

1 year

@ylecun IT has had a modest impact on US productivity. Electricity was way more impactful. Will AI match the impact of electricity? Source:

3

5

34

Serafim Batzoglou

@s_batzoglou

2 months

@GaryMarcus Haaaahahha you got me fooled for like 10 seconds. Became obvious at the “open” part

0

34

Serafim Batzoglou

@s_batzoglou

1 year

@The_Matt_Brooks @sama It is shrewdly expressed. In other words “it is important to ban open AI projects above a capability threshold”. Self-serving.

1

0

33

Serafim Batzoglou

@s_batzoglou

1 year

@tegmark @elonmusk I disagree, halting progress isn't realistic or beneficial. Let's focus on tech solutions to manage potential risks instead

2

0

32

Serafim Batzoglou

@s_batzoglou

5 months

@ZubyMusic Next step is TV commercials

4

0

32

Serafim Batzoglou

@s_batzoglou

5 months

@BillAckman @Harvard Excellent letter

0

31

Serafim Batzoglou

@s_batzoglou

1 year

@erikphoel This is a disastrous decision. The US's merit based system has been key to its success in science and business. I also climbed that ladder, starting from right after high school in Greece. This ladder is now being dismantled.

1

3

30

Serafim Batzoglou

@s_batzoglou

6 months

@AndrewYNg @sama @gdb @miramurati It is tragic that a bad BOD decision can prove so catastrophic. Why is the BOD not resigning yet?

2

0

31

Serafim Batzoglou

@s_batzoglou

1 year

@geoffreyhinton Typical journalists. However, I agree with the second sentence but not with the first one.

3

0

31

Serafim Batzoglou

@s_batzoglou

1 month

@LongFormMath The real line is not real

2

32

Serafim Batzoglou

@s_batzoglou

1 year

@WSJ So much for the 6 month moratorium

3

1

30

Serafim Batzoglou

@s_batzoglou

1 year

@WallStreetSilv Hard to ignore given the moves of China, Russia, Saudi Arabia and a number of other countries

1

0

31

Serafim Batzoglou

@s_batzoglou

2 months

@pmddomingos Time to write some new EU regulation that will mandate the increase of the value of the EU tech sector

2

1

31

Serafim Batzoglou

@s_batzoglou

1 year

@pmddomingos @lexfridman Precisely right. You may want to add print(“hello world”) as a first line, just to be polite

0

1

28

Serafim Batzoglou

@s_batzoglou

17 days

So good to see @manoliskellis

1

3

30

Serafim Batzoglou

@s_batzoglou

11 months

@JuddLegum @elonmusk Nothing is perfect but I think Twitter is fine

12

0

29

Serafim Batzoglou

@s_batzoglou

1 year

@ylecun Exactly right. Thank you for being a relentless proponent of open AI (not “OpenAI”) and a realist

1

0

28

Serafim Batzoglou

@s_batzoglou

1 year

@GaryMarcus Sentience won’t just pop up out of nowhere. No matter how impressive the results are, we fully understand to date how AIs work

4

0

26

Serafim Batzoglou

@s_batzoglou

1 year

@NathanHarden I like the idea of each Stanford student having a personal butler. Probably better use of the resources for educational purposes

0

25

Serafim Batzoglou

@s_batzoglou

1 year

@daniel_eth The author is a Dune fan for sure. Whichever country issues an AI ban goes to the bottom of the international competitiveness pile within a decade. I definitely move out

2

0

26

Serafim Batzoglou

@s_batzoglou

8 months

And the forum will not include any academics working on AI @GaryMarcus @chrmanning @BoWang87 @anshulkundaje

DogeDesigner

@cb_doge

8 months

Elon Musk and Mark Zuckerberg will brief U.S. senators on artificial intelligence today. The forum will also feature Sam Altman, Sundar Pichai and Bill Gates.

539

547

4K

1

5

28

Serafim Batzoglou

@s_batzoglou

10 months

@ScottAdamsSays Sounds like an opportunity for a new insurance company

1

25

Serafim Batzoglou

@s_batzoglou

5 months

@pmarca I read it, good book but can do better. Advanced Linear Algebra by Roman is both more advanced and more intuitive. Way way better imo.

2

1

26

Serafim Batzoglou

@s_batzoglou

1 year

@lexfridman This is also my key disagreement with @manoliskellis in his beautiful latest interview with you. AI will soon be superb in arguing it is a person but will still lack sentience. We don’t have an effective way to tell a sentient from a non sentient being.

4

1

25

Serafim Batzoglou

@s_batzoglou

7 months

@soumithchintala We should all sign this open letter in favor of open source AI

Joint Statement on AI Safety and Openness

We are at a critical juncture in AI governance. To mitigate current and future harms from AI systems, we need to embrace openness, transparency, and broad access.

open.mozilla.org

0

7

27

Serafim Batzoglou

@s_batzoglou

6 months

@Philip_Goff It is funny how many people try to dismiss philosophy in their replies to you, and do so by stating a fringe philosophical position with no argument to support it

0

27

Serafim Batzoglou

@s_batzoglou

6 months

@GadSaad The Matrix John Wick 1 and 2

0

1

27

Serafim Batzoglou

@s_batzoglou

1 year

@DJ_Mankowitz @DeepMind @Nature Congratulations! One would think that sorting of just 3 numbers couldn’t be optimized any further

0

27

Serafim Batzoglou

@s_batzoglou

3 months

@drpablocorral @JohnKastelein @SBakerMD This is super healthy, not dangerous at all

0

27

Serafim Batzoglou

@s_batzoglou

1 year

The voices against open AI and for premature regulation are too loud and unrepresentative. The majority of AI researchers and colleagues are in favor openness and see the tremendous benefits of AI @manoliskellis @GaryMarcus @chrmanning @BoWang87

AI is incredibly exciting; call to pause developments can be counter-productive: Sebastian Thrun

Elon Musk, Apple co-founder Steve Wozniak, experts, and industry leaders are demanding a pause on developments in artificial intelligence to put in place safety protocols, but celebrated AI pioneer...

economictimes.indiatimes.com

3

8

27

Serafim Batzoglou

@s_batzoglou

1 year

@JosephJacks_ It is truly hilariously transparent. OpenAI is against open AI that has capability higher than 2 levels below their latest GPT

1

5

26

Serafim Batzoglou

@s_batzoglou

4 months

AlphaGeometry is brilliant. The method seems super challenging to extend beyond geometry, but still represents a big advance. The most important factor for its success is perhaps the synthetic data generation method for training. Given a geometric diagram, a symbolic deduction

0

4

25

Serafim Batzoglou

@s_batzoglou

1 year

@BrianRoemmele I asked GPT4 to write a little essay on quantum consciousness, and the result is quite good:

2

6

23

Serafim Batzoglou

@s_batzoglou

1 year

“The AI revolution in medicine”, just in time for weekend reading @zakkohane @atulbutte @sama @GaryMarcus @hajirasouliha

2

3

24

Serafim Batzoglou

@s_batzoglou

7 months

@ScottAdamsSays Great! I never eat any bread or starches. 4 months ago I stopped having any sweeteners and any protein shakes. Replaced with Greek yogurt and eggs. I’ve had instantaneous and remarkable improvement all around. Stick to real foods and low carb.

4

1

25

Serafim Batzoglou

@s_batzoglou

5 months

@BrendanCarrFCC @ncri_io I would be surprised if this wasn’t the case

0

25

Serafim Batzoglou

@s_batzoglou

1 year

@GaryMarcus This is a terrible proposal and I hope nothing comes out of it

2

1

25

Serafim Batzoglou

@s_batzoglou

6 months

@ylecun @geoffreyhinton Much as I admire @geoffreyhinton ’s contributions, I am still waiting for a convincing argument for AI doom risk. In an abstract intellectual sense there is some plausibility, but that’s about it. And even suspending disbelief and accepting the doom argument hypothetically, the

4

0

20