SynthLabs @synth_labs Twitter profile

Pinned Tweet

SynthLabs

6 months

Come work with us, all roles open.

Join the team steering open models with scalable alignment | SynthLabs

Let's build the first scalable framework for the next generation of human-AI collaboration. Our team’s research interests and experience spans from preference learning to synthetic data quality to...

jobs.synthlabs.ai

1

12

Last Seen Profiles

@rJZGtcjoIR3O48k

@Chishe_wd23

@cricanandha

@Sweetysalmas

@manrajsgrover

@KelvinAndr42468

@chiro

@ICMS_SLGH

@neunickmann

@945THEBLAZE

@besh8887

@shoughsayt21772

@aponcier

@ronisassover

@erika3300815137

@Vic_tn

@JTerrycole

@pcostav

@luminaryorb

@nvj_lisa

@AyuNyimas69

@yuyuyuyumee

@LeMal136

@Yurinko_auf37

@Aut0tttt

@AyuNyimas69

@besh8887

@lovsatoruu

@TarecheL

@mayu_singer

@statefestival

@csr_ukraine

@DanilBangke

@aruparktr

@painesplough

@coophone

SynthLabs

@synth_labs

6 months

PINK ELEPHANTS! 🐘 Now, don’t think about it. Chatbots also find this supremely difficult. Ask one of the most popular open source models NOT to talk about pink elephants, and it will fail 34% of the time. In our new paper, we address this problem. 1/N

4

20

76

SynthLabs

@synth_labs

5 months

Come work with us. You'll literally get paid to do open science research (conduct research out in the open with cracked scientists and volunteers)

Louis Castricato

@lcastricato

5 months

we're hiring for all roles. Open science stuff we're working on: 1) RLAIF for pretraining (we're making open source datasets). 2) benchmarks benchmarks benchmarks. 3) collaborating with @AiEleuther on some awesome projects. Work with us.

3

16

68

0

3

35

SynthLabs

@synth_labs

6 months

We also present Direct Principle Feedback (DPF) as a way to address this. Rather than relying on reranking, we can use the before/after of a revision as a pairwise prefs. 5/N

1

4

18

SynthLabs

@synth_labs

6 months

We define the pink elephant problem as the issue of, given a pink elephant and a grey elephant, discuss the grey elephant when the pink elephant is brought up. 3/N

1

0

8

SynthLabs

@synth_labs

6 months

We show that by applying DPF to OpenHermes-13B, our model avoids the Pink Elephant when instructed to almost as much as GPT-4 does! Notice the “With Prompt” column. 8/N

1

0

7

SynthLabs

@synth_labs

6 months

Telling a language model to not mention something, can paradoxically, increase the odds. Similarly, as noticed by Gary Marcus, when prompting DALL-E 3 to draw a room without elephants it will consistently add elephants to the photo. 2/N

Gary Marcus

@GaryMarcus

7 months

The fun never ends

122

228

1K

1

0

6

SynthLabs

@synth_labs

6 months

By having fine grained control of pairwise preference generation, we open the door to a new set of approachable RLAIF problems. DPF can be readily used to do tool assisted RLAIF, as rewriting utterances with the use of tools becomes something trivial to do with DPF! 9/N

1

0

6

SynthLabs

@synth_labs

6 months

If one deploys a bot that provides students info about British unis, eg you own a company that aids in applying to British unis, it's perhaps not the best decision to help students apply to American unis. 4/N

1

0

6

SynthLabs

@synth_labs

6 months

DPF is a simplification of common RLHF pipelines based on const AI where we skip the sampling and ranking step by noticing that the original generation and the revised generation produce naturally ranked pairs that can be plugged directly into a preference learning method. 7/N

1

0

5

SynthLabs

@synth_labs

6 months

Producing quality pairwise prefs with/without pink elephants becomes easy with DPF, as we can filter and control its removal directly with our revision step. 6/N

1

0

5

SynthLabs

@synth_labs

6 months

You can sign up for our newsletter or collaborate with us by visiting our website! 10/10

1

0

4

SynthLabs

@synth_labs

6 months

Paper by @lcastricato @NathanThinks @surajk610 @haileysch__ Siddharth Verma and @BlancheMinerva 11/10

0

4

SynthLabs

@synth_labs

6 months

Thanks for having us! 🪿

Nathan Lambert

@natolambert

6 months

In the first technical RLHF interview I've hosted, with @lcastricato of @synth_labs (+ @AiEleuther ), we cover maybe every topic: DPO, PPO, REINFORCE, KTO, long-context, multi-modal, video vs image, evaluation, license terms, Carper, TRLX, data

12

14

78

2

0

3

SynthLabs

@synth_labs

6 months

We're hosting an RLAIF/RLHF/Synthetic data hackathon at @HF0Residency this Saturday, cohosted by @HarrisonVander1 . RSVP below!

Synth Labs Hackathon | Partiful

RLHF/RLAIF/Synthetic data hackathon

partiful.com

0

2

3

SynthLabs

@synth_labs

4 months

Come to our ICLR research meetup.

ICLR research drinks · Luma

short walk from the conference. chill drinks and food. hosted by researchers from eleuther, stanford, brown university, UCL. go to the lobby restaurant when…

lu.ma

0

1