[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.] #  @iScienceLuvr Tanishq Mathew Abraham, Ph.D. Tanishq Mathew Abraham, Ph.D. posts on X about 6969, gpu, reinforcement learning, mae the most. They currently have XXXXXX followers and XXX posts still getting attention that total XXXXXXX engagements in the last XX hours. ### Engagements: XXXXXXX [#](/creator/twitter::441465751/interactions)  - X Week XXXXXXX +165% - X Month XXXXXXXXX +144% - X Months XXXXXXXXX -XX% - X Year XXXXXXXXXX +101% ### Mentions: XX [#](/creator/twitter::441465751/posts_active)  - X Month XX +6.30% - X Months XXX -XX% - X Year XXX +115% ### Followers: XXXXXX [#](/creator/twitter::441465751/followers)  - X Week XXXXXX +0.33% - X Month XXXXXX +1.90% - X Months XXXXXX +8.40% - X Year XXXXXX +33% ### CreatorRank: XXXXXX [#](/creator/twitter::441465751/influencer_rank)  ### Social Influence [#](/creator/twitter::441465751/influence) --- **Social category influence** [technology brands](/list/technology-brands) **Social topic influence** [6969](/topic/6969), [gpu](/topic/gpu), [reinforcement learning](/topic/reinforcement-learning), [mae](/topic/mae), [applications](/topic/applications), [agi](/topic/agi), [$922m](/topic/$922m), [android](/topic/android), [token](/topic/token), [generated](/topic/generated) ### Top Social Posts [#](/creator/twitter::441465751/posts) --- Top posts by engagements in the last XX hours "QeRL: Beyond Efficiency -- Quantization-enhanced Reinforcement Learning for LLMs "this is the first framework to enable RL training of a 32B LLM on a single H100 80GB GPU" "QeRL addresses these issues by combining NVFP4 quantization with Low-Rank Adaptation (LoRA) accelerating rollout phase of RL while reducing memory overhead."" [X Link](https://x.com/iScienceLuvr/status/1978046212621373719) [@iScienceLuvr](/creator/x/iScienceLuvr) 2025-10-14T10:32Z 82.5K followers, 6967 engagements "Diffusion Transformers with Representation Autoencoders "Most DiTs continue to rely on the original VAE encoder" "In this work we explore replacing the VAE with pretrained representation encoders (e.g. DINO SigLIP MAE) paired with trained decoders forming what we term Representation Autoencoders (RAEs)." "Using a DiT variant equipped with a lightweight wide DDT head we achieve strong image generation results on ImageNet: XXXX FID at 256x256 (no guidance) and XXXX at both 256x256 and 512x512 (with guidance)."" [X Link](https://x.com/iScienceLuvr/status/1978053094769615296) [@iScienceLuvr](/creator/x/iScienceLuvr) 2025-10-14T10:59Z 82.5K followers, 21.1K engagements "We've released our first @SophontAI paper 🔥 We're bullish on the potential foundation models to help unlock novel clinical applications for brain & mental health. So we working on training better neuroimaging foundation models. We develop a novel approach for training fMRI foundation models by projecting the fMRI data onto a flattened map of the brain cortex and training a spatiotemporal MAE on that data. We observe a nice dataset scaling law and strong performance on downstream tasks but there's lot more work to do. So we're opening up the project for contributions join our @MedARC_AI" [X Link](https://x.com/iScienceLuvr/status/1978896050732847285) [@iScienceLuvr](/creator/x/iScienceLuvr) 2025-10-16T18:49Z 82.5K followers, 52.3K engagements "I don't like this idea that we should solve AGI first and then AGI will solve everything else. When you can make a difference now why not do that I've seen a lot of people not want to work on medical AI for this reason: "AGI will cure cancer why bother now" But I think people fail to realize the impact biomedical foundation models can have in the next few years from accelerating drug development to better to making it easier to navigate the healthcare system to choosing better treatments for patients. From freeing up doctor's time to keeping patients informed. These applications can change" [X Link](https://x.com/iScienceLuvr/status/1980715994210529306) [@iScienceLuvr](/creator/x/iScienceLuvr) 2025-10-21T19:21Z 82.5K followers, 3429 engagements "Excited to announce that @SophontAI has raised $9.22M in combined pre-seed+seed rounds 🚀🔥 Led by @KindredVentures with participation from @delphi_ventures @upfrontvc @AICONIC_VC also @jeffdean @logankilpatrick @ClementDelangue (via Factorial Capital) @l2k & others" [X Link](https://x.com/iScienceLuvr/status/1966202358519763196) [@iScienceLuvr](/creator/x/iScienceLuvr) 2025-09-11T18:09Z 82.5K followers, 159K engagements "SimpleFold: Folding Proteins is Simpler than You Think "we introduce SimpleFold the first flow-matching based protein folding model that solely uses general purpose transformer blocks. Protein folding models typically employ computationally expensive modules involving triangular updates explicit pair representations or multiple training objectives curated for this specific domain. Instead SimpleFold employs standard transformer blocks with adaptive layers and is trained via a generative flow-matching objective with an additional structural term."" [X Link](https://x.com/iScienceLuvr/status/1970787581248905454) [@iScienceLuvr](/creator/x/iScienceLuvr) 2025-09-24T09:49Z 82.5K followers, 284.5K engagements "I unfortunately don't know much about mobile app and desktop software dev so I'm genuinely asking: WHY DO ALL COMPANIES RELEASE iOS/macOS PRODUCTS FIRST/ONLY How hard is it to release Android and Windows versions of your product Recent example is the Sora app which is only released for iOS with Android coming later. This means Android users like myself cannot do Cameos and are missing out on the internet moment. I've seen many other products that are iOS/macOS-only and it's really frustrating that I cannot use them at all. Why would you purposefully cut out like half of the population who can" [X Link](https://x.com/iScienceLuvr/status/1973460927942214111) [@iScienceLuvr](/creator/x/iScienceLuvr) 2025-10-01T18:52Z 82.5K followers, 461.9K engagements "Attention Sinks and Compression Valleys in LLMs are Two Sides of the Same Coin "Through experiments across several models (410M120B parameters) we confirm that when the beginning-of-sequence token develops extreme activation norms in the middle layers both compression valleys and attention sinks emerge simultaneously." "we posit that Transformer-based LLMs process tokens in three distinct phases: (1) broad mixing in the early layers (2) compressed computation with limited mixing in the middle layers and (3) selective refinement in the late layers."" [X Link](https://x.com/iScienceLuvr/status/1976235853853909048) [@iScienceLuvr](/creator/x/iScienceLuvr) 2025-10-09T10:38Z 82.5K followers, 24.6K engagements "Agent Learning via Early Experience "training agents from experience data with reinforcement learning remains difficult in many environments which either lack verifiable rewards (e.g. websites) or require inefficient long-horizon rollouts (e.g. multi-turn tool use)." "We address this limitation with a middle-ground paradigm we call early experience: interaction data generated by the agents own actions where the resulting future states serve as supervision without reward signals. Within this paradigm we study two strategies of using such data: (1) Implicit world modeling which uses collected" [X Link](https://x.com/iScienceLuvr/status/1976588352746107204) [@iScienceLuvr](/creator/x/iScienceLuvr) 2025-10-10T09:59Z 82.5K followers, 28.5K engagements "NExT-OMNI: Towards Any-to-Any Omnimodal Foundation Models with Discrete Flow Matching "In this work we introduce NExT-OMNI an open-source omnimodal foundation model that achieves unified modeling through discrete flow paradigms." "Trained on large-scale interleaved text image video and audio data NExT-OMNI delivers competitive performance on multimodal generation and understanding benchmarks while outperforming prior unified models in multi-turn multimodal interaction and cross-modal retrieval"" [X Link](https://x.com/iScienceLuvr/status/1978788541770695145) [@iScienceLuvr](/creator/x/iScienceLuvr) 2025-10-16T11:42Z 82.5K followers, 8690 engagements "This ad is on the back of an issue of IEEE Spectrum. And I want to ask: do people actually use MATLAB for AI" [X Link](https://x.com/iScienceLuvr/status/1979691790732333220) [@iScienceLuvr](/creator/x/iScienceLuvr) 2025-10-18T23:31Z 82.5K followers, 8772 engagements "the thing about old AI art was that it used to have distinct styles. now a lot of the AI art is quite bland imo" [X Link](https://x.com/iScienceLuvr/status/1980115673864106441) [@iScienceLuvr](/creator/x/iScienceLuvr) 2025-10-20T03:35Z 82.5K followers, 12.3K engagements "DeepSeek released an OCR model today. Their motivation is really interesting: they want to use visual modality as an efficient compression medium for textual information and use this to solve long-context challenges in LLMs. Of course they are using it to get more training data for their models as well. "DeepSeek-OCR can generate training data for LLMs/VLMs at a scale of 200k+ pages per day (a single A100-40G)."" [X Link](https://x.com/iScienceLuvr/status/1980247935700066468) [@iScienceLuvr](/creator/x/iScienceLuvr) 2025-10-20T12:21Z 82.5K followers, 265.2K engagements "Maybe computer vision is all you need lol To be fair that's how humans work it isn't like language is a separate sense lol" [X Link](https://x.com/iScienceLuvr/status/1980445128553804145) [@iScienceLuvr](/creator/x/iScienceLuvr) 2025-10-21T01:24Z 82.5K followers, 172.8K engagements "Fwiw while this is technically true does it matter Language models are sampling over some vocabulary distribution and so as long as there are no tokens with zero probability any sequence of text can technically be produced by the base model just at really low probabilities. So technically everything is a capability of the base model You just need enough rollouts to see it. Then RL enables you to increase the probability density of the desired rollouts. If you have infinite monkeys to write Shakespeare then you can RL them to be Shakespeare experts" [X Link](https://x.com/iScienceLuvr/status/1980497393209725271) [@iScienceLuvr](/creator/x/iScienceLuvr) 2025-10-21T04:52Z 82.5K followers, 20.1K engagements "@pritmish well basically yeah lol" [X Link](https://x.com/iScienceLuvr/status/1980560930837799156) [@iScienceLuvr](/creator/x/iScienceLuvr) 2025-10-21T09:04Z 82.5K followers, XXX engagements "Today I started getting these completely garbage bot responses. I hate it my notifications are spammed with these. @nikitabier please fix" [X Link](https://x.com/iScienceLuvr/status/1980792022190743823) [@iScienceLuvr](/creator/x/iScienceLuvr) 2025-10-22T00:23Z 82.5K followers, 2103 engagements "This is actually kinda sad. Using ChatGPT in WhatsApp was a fun hack to use ChatGPT on flights for free" [X Link](https://x.com/iScienceLuvr/status/1980795940501090384) [@iScienceLuvr](/creator/x/iScienceLuvr) 2025-10-22T00:38Z 82.5K followers, 20.9K engagements
[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]
@iScienceLuvr Tanishq Mathew Abraham, Ph.D.
Tanishq Mathew Abraham, Ph.D. posts on X about 6969, gpu, reinforcement learning, mae the most. They currently have XXXXXX followers and XXX posts still getting attention that total XXXXXXX engagements in the last XX hours.
Engagements: XXXXXXX #
- X Week XXXXXXX +165%
- X Month XXXXXXXXX +144%
- X Months XXXXXXXXX -XX%
- X Year XXXXXXXXXX +101%
Mentions: XX #
- X Month XX +6.30%
- X Months XXX -XX%
- X Year XXX +115%
Followers: XXXXXX #
- X Week XXXXXX +0.33%
- X Month XXXXXX +1.90%
- X Months XXXXXX +8.40%
- X Year XXXXXX +33%
CreatorRank: XXXXXX #
Social Influence #
Social category influence technology brands
Social topic influence 6969, gpu, reinforcement learning, mae, applications, agi, $922m, android, token, generated
Top Social Posts #
Top posts by engagements in the last XX hours
"QeRL: Beyond Efficiency -- Quantization-enhanced Reinforcement Learning for LLMs "this is the first framework to enable RL training of a 32B LLM on a single H100 80GB GPU" "QeRL addresses these issues by combining NVFP4 quantization with Low-Rank Adaptation (LoRA) accelerating rollout phase of RL while reducing memory overhead.""
X Link @iScienceLuvr 2025-10-14T10:32Z 82.5K followers, 6967 engagements
"Diffusion Transformers with Representation Autoencoders "Most DiTs continue to rely on the original VAE encoder" "In this work we explore replacing the VAE with pretrained representation encoders (e.g. DINO SigLIP MAE) paired with trained decoders forming what we term Representation Autoencoders (RAEs)." "Using a DiT variant equipped with a lightweight wide DDT head we achieve strong image generation results on ImageNet: XXXX FID at 256x256 (no guidance) and XXXX at both 256x256 and 512x512 (with guidance).""
X Link @iScienceLuvr 2025-10-14T10:59Z 82.5K followers, 21.1K engagements
"We've released our first @SophontAI paper 🔥 We're bullish on the potential foundation models to help unlock novel clinical applications for brain & mental health. So we working on training better neuroimaging foundation models. We develop a novel approach for training fMRI foundation models by projecting the fMRI data onto a flattened map of the brain cortex and training a spatiotemporal MAE on that data. We observe a nice dataset scaling law and strong performance on downstream tasks but there's lot more work to do. So we're opening up the project for contributions join our @MedARC_AI"
X Link @iScienceLuvr 2025-10-16T18:49Z 82.5K followers, 52.3K engagements
"I don't like this idea that we should solve AGI first and then AGI will solve everything else. When you can make a difference now why not do that I've seen a lot of people not want to work on medical AI for this reason: "AGI will cure cancer why bother now" But I think people fail to realize the impact biomedical foundation models can have in the next few years from accelerating drug development to better to making it easier to navigate the healthcare system to choosing better treatments for patients. From freeing up doctor's time to keeping patients informed. These applications can change"
X Link @iScienceLuvr 2025-10-21T19:21Z 82.5K followers, 3429 engagements
"Excited to announce that @SophontAI has raised $9.22M in combined pre-seed+seed rounds 🚀🔥 Led by @KindredVentures with participation from @delphi_ventures @upfrontvc @AICONIC_VC also @jeffdean @logankilpatrick @ClementDelangue (via Factorial Capital) @l2k & others"
X Link @iScienceLuvr 2025-09-11T18:09Z 82.5K followers, 159K engagements
"SimpleFold: Folding Proteins is Simpler than You Think "we introduce SimpleFold the first flow-matching based protein folding model that solely uses general purpose transformer blocks. Protein folding models typically employ computationally expensive modules involving triangular updates explicit pair representations or multiple training objectives curated for this specific domain. Instead SimpleFold employs standard transformer blocks with adaptive layers and is trained via a generative flow-matching objective with an additional structural term.""
X Link @iScienceLuvr 2025-09-24T09:49Z 82.5K followers, 284.5K engagements
"I unfortunately don't know much about mobile app and desktop software dev so I'm genuinely asking: WHY DO ALL COMPANIES RELEASE iOS/macOS PRODUCTS FIRST/ONLY How hard is it to release Android and Windows versions of your product Recent example is the Sora app which is only released for iOS with Android coming later. This means Android users like myself cannot do Cameos and are missing out on the internet moment. I've seen many other products that are iOS/macOS-only and it's really frustrating that I cannot use them at all. Why would you purposefully cut out like half of the population who can"
X Link @iScienceLuvr 2025-10-01T18:52Z 82.5K followers, 461.9K engagements
"Attention Sinks and Compression Valleys in LLMs are Two Sides of the Same Coin "Through experiments across several models (410M120B parameters) we confirm that when the beginning-of-sequence token develops extreme activation norms in the middle layers both compression valleys and attention sinks emerge simultaneously." "we posit that Transformer-based LLMs process tokens in three distinct phases: (1) broad mixing in the early layers (2) compressed computation with limited mixing in the middle layers and (3) selective refinement in the late layers.""
X Link @iScienceLuvr 2025-10-09T10:38Z 82.5K followers, 24.6K engagements
"Agent Learning via Early Experience "training agents from experience data with reinforcement learning remains difficult in many environments which either lack verifiable rewards (e.g. websites) or require inefficient long-horizon rollouts (e.g. multi-turn tool use)." "We address this limitation with a middle-ground paradigm we call early experience: interaction data generated by the agents own actions where the resulting future states serve as supervision without reward signals. Within this paradigm we study two strategies of using such data: (1) Implicit world modeling which uses collected"
X Link @iScienceLuvr 2025-10-10T09:59Z 82.5K followers, 28.5K engagements
"NExT-OMNI: Towards Any-to-Any Omnimodal Foundation Models with Discrete Flow Matching "In this work we introduce NExT-OMNI an open-source omnimodal foundation model that achieves unified modeling through discrete flow paradigms." "Trained on large-scale interleaved text image video and audio data NExT-OMNI delivers competitive performance on multimodal generation and understanding benchmarks while outperforming prior unified models in multi-turn multimodal interaction and cross-modal retrieval""
X Link @iScienceLuvr 2025-10-16T11:42Z 82.5K followers, 8690 engagements
"This ad is on the back of an issue of IEEE Spectrum. And I want to ask: do people actually use MATLAB for AI"
X Link @iScienceLuvr 2025-10-18T23:31Z 82.5K followers, 8772 engagements
"the thing about old AI art was that it used to have distinct styles. now a lot of the AI art is quite bland imo"
X Link @iScienceLuvr 2025-10-20T03:35Z 82.5K followers, 12.3K engagements
"DeepSeek released an OCR model today. Their motivation is really interesting: they want to use visual modality as an efficient compression medium for textual information and use this to solve long-context challenges in LLMs. Of course they are using it to get more training data for their models as well. "DeepSeek-OCR can generate training data for LLMs/VLMs at a scale of 200k+ pages per day (a single A100-40G).""
X Link @iScienceLuvr 2025-10-20T12:21Z 82.5K followers, 265.2K engagements
"Maybe computer vision is all you need lol To be fair that's how humans work it isn't like language is a separate sense lol"
X Link @iScienceLuvr 2025-10-21T01:24Z 82.5K followers, 172.8K engagements
"Fwiw while this is technically true does it matter Language models are sampling over some vocabulary distribution and so as long as there are no tokens with zero probability any sequence of text can technically be produced by the base model just at really low probabilities. So technically everything is a capability of the base model You just need enough rollouts to see it. Then RL enables you to increase the probability density of the desired rollouts. If you have infinite monkeys to write Shakespeare then you can RL them to be Shakespeare experts"
X Link @iScienceLuvr 2025-10-21T04:52Z 82.5K followers, 20.1K engagements
"@pritmish well basically yeah lol"
X Link @iScienceLuvr 2025-10-21T09:04Z 82.5K followers, XXX engagements
"Today I started getting these completely garbage bot responses. I hate it my notifications are spammed with these. @nikitabier please fix"
X Link @iScienceLuvr 2025-10-22T00:23Z 82.5K followers, 2103 engagements
"This is actually kinda sad. Using ChatGPT in WhatsApp was a fun hack to use ChatGPT on flights for free"
X Link @iScienceLuvr 2025-10-22T00:38Z 82.5K followers, 20.9K engagements