#thinkless
LightNews LightNews
beta8088proto.protogen.club
absolutely thinkless critter
mythiccvixen.bsky.social 🦊 MythiccVixen 🔞 @mythiccvixen.bsky.social · 11d
Still recovering but:

You know how playing dogs charge & pounce something, putting their full weight & momentum behind their paws?

This 80+ pound ding-a-ling did that to my bad hand.

Still have extra pain & stiffness but getting better.

She’s fine. Didn’t get yelled at. Just sad playtime ended.
Large black dog lays on her side, belly away from you, but neck at a right angle so she can look at you with a piteous expression. Head still on the floor, her chin rests against the side of her bed as does one white-footed paw next to her cheek.
October 30, 2026 at 11:57 PM
reel360news.bsky.social
Ford and Think Less put community ahead of product in “La Caravana,” sending the Ventura Twins 2,700 miles from Miami to LA in an F-150 Lobo to explore Latino food, lowrider culture and Pacoima’s Mural Mile. #Ford #LaCaravana #ThinkLess #BrandedContent #Reel360
Ford and Think Less put Latino communities in driver’s seat
Ford Motor Company and Salt Lake City creative and technology studio Think Less hit the road for “La Caravana,” a three-part social series that follows creators Brayan and Brando Ventura - better known as the Ventura Twins - on a 2,700-mile journey from Miami to Los Angeles in an F-150 Lobo. But the truck isn’t really the star. The people are.
reel360.com
November 3, 2026 at 8:26 PM
steelcrustacean.bsky.social
negative five thoughts floating within her cranium, the thinkless one
October 23, 2026 at 5:39 PM
beta8088proto.protogen.club
thinkless gaze
10n.sodadr.one 10N Protogen 🔜 ANE27 @10n.sodadr.one · 28d
She's a silly creature
White and brown cat looking at the camera
October 14, 2026 at 3:53 AM
senselessfool.bsky.social
i wont steal this one because i'm opting for some responding without storing action here at our beautiful 5:20 am where i am thinkless and my fingers are doing all the thinking.

but dont you worry i will steal away when so inclined as i often have done :)

caw!

caw!

caw!
October 5, 2026 at 12:21 PM
prettylilbitch.bsky.social
I don’t show faces — I reveal fragments. The rest is for you to imagine.

Welcome to the art of becoming your own canvas.

#photoedits #portraits #photography #filters #photoart #like #follow #share #confidenceiskey #thinkless #domore
November 9, 2025 at 10:17 AM
paper.bsky.social
2505.13379
拡張された思考連鎖推論が可能な推論言語モデルは、複雑な論理推論を必要とするタスクで目覚ましい性能を発揮している。しかし、すべてのクエリに精巧な推論を適用すると、特に多くの問題が簡単な解を持つ場合に、計算効率が大幅に低下することが多い。これは、未解決の疑問の動機となる：LLMはいつ考えるべ...
拡張された思考連鎖推論が可能な推論言語モデルは、複雑な論理推論を必要とするタスクで目覚ましい性能を発揮している。 しかし、すべてのクエリに精巧な推論を適用すると、特に多くの問題が簡単な解を持つ場合に、計算効率が大幅に低下することが多い。 これは、未解決の疑問の動機となる：LLMはいつ考えるべきかを学べるのか？ この問題に答えるために、我々はThinklessを提案する。Thinklessは、タスクの複雑さとモデルの能力の両方に基づいて、LLMが短文推論と長文推論を適応的に選択できるようにする学習可能なフレームワークである。 Thinklessは強化学習パラダイムで学習され、簡潔な応答には<short>、詳細な推論には<think>という2つのコントロール・トークンを用いる。 我々の手法の核となるのは、ハイブリッド推論の学習目的を2つの要素に分解する分離グループ相対政策最適化(DeGRPO)アルゴリズムである：(1)推論モードの選択を支配する制御トークン損失、(2)生成された答えの精度を向上させる応答損失。 この分離された定式化により、各目的の貢献度をきめ細かく制御することが可能になり、トレーニングを安定させ、バニラGRPOで観察される崩壊を効果的に防ぐことができる。 経験的に、Minerva Algebra、MATH-500、GSM8Kなどのいくつかのベンチマークにおいて、Thinklessは長鎖思考の使用量を50%から90%削減することができ、推論言語モデルの効率を大幅に改善することができる。 コードはhttps://github.com/VainF/Thinkless。
June 1, 2025 at 12:08 AM
paper.bsky.social
Links: abs, pdf
Search: Bluesky, Twitter, Reddit, Hacker News, Hugging Face, alphaXiv
Thinkless: LLM Learns When to Think
Reasoning Language Models, capable of extended chain-of-thought reasoning, have demonstrated remarkable performance on tasks requiring complex logical inference. However, applying elaborate reasoning ...
arxiv.org
June 1, 2025 at 12:08 AM
paper.bsky.social
(3/3) 32 Likes, 0 Comments, 23 May 2025, Reddit
From the machinelearningnews community on Reddit: Researchers from the National University of Singapore Introduce ‘Thinkless,’ an Adaptive Framework that Reduces Unnecessary Reasoning by up to 90% Usi...
Explore this post and more from the machinelearningnews community
redd.it
June 1, 2025 at 12:08 AM
paper.bsky.social
(2/3) 47 Likes, 2 Comments, 20 May 2025, Hugging Face
Paper page - Thinkless: LLM Learns When to Think
Join the discussion on this paper page
huggingface.co
June 1, 2025 at 12:08 AM
paper.bsky.social
(1/3) 489 Likes, 0 Comments, 20 May 2025, alphaXiv
Thinkless: LLM Learns When to Think | alphaXiv
View 1 comments: Could there ever be a dataset and set of instructions perhaps that applies some formal logic or symbolic logic to non-scientific domains, such that it would be possible to separate si...
www.alphaxiv.org
June 1, 2025 at 12:08 AM
paper.bsky.social
[16/30] 568 Likes, 2 Comments, 3 Posts
2505.13379, cs․CL | cs․AI, 19 May 2025

🆕Thinkless: LLM Learns When to Think

Gongfan Fang, Xinyin Ma, Xinchao Wang
Reasoning Language Models, capable of extended chain-of-thought reasoning, have demonstrated remarkable performance on tasks requiring complex logical inference. However, applying elaborate reasoning for all queries often results in substantial computational inefficiencies, particularly when many problems admit straightforward solutions. This motivates an open question: Can LLMs learn when to think? To answer this, we propose Thinkless, a learnable framework that empowers an LLM to adaptively select between short-form and long-form reasoning, based on both task complexity and the model's ability. Thinkless is trained under a reinforcement learning paradigm and employs two control tokens, <short> for concise responses and <think> for detailed reasoning. At the core of our method is a Decoupled Group Relative Policy Optimization (DeGRPO) algorithm, which decomposes the learning objective of hybrid reasoning into two components: (1) a control token loss that governs the selection of the reasoning mode, and (2) a response loss that improves the accuracy of the generated answers. This decoupled formulation enables fine-grained control over the contributions of each objective, stabilizing training and effectively preventing collapse observed in vanilla GRPO. Empirically, on several benchmarks such as Minerva Algebra, MATH-500, and GSM8K, Thinkless is able to reduce the usage of long-chain thinking by 50% - 90%, significantly improving the efficiency of Reasoning Language Models. The code is available at https://github.com/VainF/Thinkless
June 1, 2025 at 12:08 AM
paper.bsky.social
2505.13379
拡張された思考連鎖推論が可能な推論言語モデルは、複雑な論理推論を必要とするタスクで目覚ましい性能を発揮している。しかし、すべてのクエリに精巧な推論を適用すると、特に多くの問題が簡単な解を持つ場合に、計算効率が大幅に低下することが多い。これは、未解決の疑問の動機となる：LLMはいつ考えるべ...
拡張された思考連鎖推論が可能な推論言語モデルは、複雑な論理推論を必要とするタスクで目覚ましい性能を発揮している。 しかし、すべてのクエリに精巧な推論を適用すると、特に多くの問題が簡単な解を持つ場合に、計算効率が大幅に低下することが多い。 これは、未解決の疑問の動機となる：LLMはいつ考えるべきかを学べるのか？ この問題に答えるために、我々はThinklessを提案する。Thinklessは、タスクの複雑さとモデルの能力の両方に基づいて、LLMが短文推論と長文推論を適応的に選択できるようにする学習可能なフレームワークである。 Thinklessは強化学習パラダイムで学習され、簡潔な応答には<short>、詳細な推論には<think>という2つのコントロール・トークンを用いる。 我々の手法の核となるのは、ハイブリッド推論の学習目的を2つの要素に分解する分離グループ相対政策最適化(DeGRPO)アルゴリズムである：(1)推論モードの選択を支配する制御トークン損失、(2)生成された答えの精度を向上させる応答損失。 この分離された定式化により、各目的の貢献度をきめ細かく制御することが可能になり、トレーニングを安定させ、バニラGRPOで観察される崩壊を効果的に防ぐことができる。 経験的に、Minerva Algebra、MATH-500、GSM8Kなどのいくつかのベンチマークにおいて、Thinklessは長鎖思考の使用量を50%から90%削減することができ、推論言語モデルの効率を大幅に改善することができる。 コードはhttps://github.com/VainF/Thinkless。
May 31, 2025 at 2:52 AM
paper.bsky.social
Links: abs, pdf
Search: Bluesky, Twitter, Reddit, Hacker News, Hugging Face, alphaXiv
Thinkless: LLM Learns When to Think
Reasoning Language Models, capable of extended chain-of-thought reasoning, have demonstrated remarkable performance on tasks requiring complex logical inference. However, applying elaborate reasoning ...
arxiv.org
May 31, 2025 at 2:52 AM
paper.bsky.social
(3/3) 34 Likes, 0 Comments, 23 May 2025, Reddit
From the machinelearningnews community on Reddit: Researchers from the National University of Singapore Introduce ‘Thinkless,’ an Adaptive Framework that Reduces Unnecessary Reasoning by up to 90% Usi...
Explore this post and more from the machinelearningnews community
redd.it
May 31, 2025 at 2:52 AM
paper.bsky.social
(2/3) 47 Likes, 2 Comments, 20 May 2025, Hugging Face
Paper page - Thinkless: LLM Learns When to Think
Join the discussion on this paper page
huggingface.co
May 31, 2025 at 2:52 AM
paper.bsky.social
(1/3) 487 Likes, 0 Comments, 20 May 2025, alphaXiv
Thinkless: LLM Learns When to Think | alphaXiv
View 1 comments: Could there ever be a dataset and set of instructions perhaps that applies some formal logic or symbolic logic to non-scientific domains, such that it would be possible to separate si...
www.alphaxiv.org
May 31, 2025 at 2:52 AM
paper.bsky.social
[15/30] 568 Likes, 2 Comments, 3 Posts
2505.13379, cs․CL | cs․AI, 19 May 2025

🆕Thinkless: LLM Learns When to Think

Gongfan Fang, Xinyin Ma, Xinchao Wang
Reasoning Language Models, capable of extended chain-of-thought reasoning, have demonstrated remarkable performance on tasks requiring complex logical inference. However, applying elaborate reasoning for all queries often results in substantial computational inefficiencies, particularly when many problems admit straightforward solutions. This motivates an open question: Can LLMs learn when to think? To answer this, we propose Thinkless, a learnable framework that empowers an LLM to adaptively select between short-form and long-form reasoning, based on both task complexity and the model's ability. Thinkless is trained under a reinforcement learning paradigm and employs two control tokens, <short> for concise responses and <think> for detailed reasoning. At the core of our method is a Decoupled Group Relative Policy Optimization (DeGRPO) algorithm, which decomposes the learning objective of hybrid reasoning into two components: (1) a control token loss that governs the selection of the reasoning mode, and (2) a response loss that improves the accuracy of the generated answers. This decoupled formulation enables fine-grained control over the contributions of each objective, stabilizing training and effectively preventing collapse observed in vanilla GRPO. Empirically, on several benchmarks such as Minerva Algebra, MATH-500, and GSM8K, Thinkless is able to reduce the usage of long-chain thinking by 50% - 90%, significantly improving the efficiency of Reasoning Language Models. The code is available at https://github.com/VainF/Thinkless
May 31, 2025 at 2:52 AM
billionerrorx.bsky.social


Thinkless thoughting
is
Mindful memory loss...

I'm pretty sure.
February 15, 2025 at 8:35 PM
skszymon.bsky.social
4/4 Co więcej, Thinkless uczy się rozróżniać proste i złożone zadania, przydzielając odpowiednie tryby odpowiedzi. To oznacza, że model jest w stanie dostosować się do różnych typów zadań i wymagań.
May 25, 2025 at 4:00 PM
skszymon.bsky.social
3/4 Wyniki są imponujące! Thinkless redukuje użycie długich odpowiedzi o 50-90% na różnych benchmarkach, takich jak AIME, Minerva Algebra i GSM8K. To oznacza, że model jest w stanie osiągnąć podobną lub lepszą dokładność niż modele bazowe, jednocześnie redukując koszty.

May 25, 2025 at 4:00 PM
skszymon.bsky.social
2/4 Thinkless wykorzystuje Reinforcement Learning do trenowania hybrydowego modelu, który adaptacyjnie wybiera tryb odpowiedzi. To oznacza, że model może zaoszczędzić czas i zasoby obliczeniowe, generując krótkie odpowiedzi, gdy jest to możliwe.

May 25, 2025 at 4:00 PM
skszymon.bsky.social
1/4 Znasz problem z modelami językowymi, które zawsze "myślą" za dużo? Nowy paper "Thinkless: LLM Learns When to Think" proponuje rozwiązanie! Model uczy się, kiedy generować krótkie odpowiedzi, a kiedy długie, w zależności od złożoności zadania.

May 25, 2025 at 4:00 PM
zyblonius.bsky.social
Me I'm really more of a thinkless actually
criminalerin.bsky.social Erin Fogg says Free Link @criminalerin.bsky.social · Jul 25
Tag yourself I'm scared on the couch with a cat
A 5 hour old The Guardian piece of some bullshit titled Four types of thinner have been identified. Which are you? And there's a problem human drawn cartoon image of just two people, one curled up on terror on the couch in headphones with a laptop and a cat, and one I'm not sure feeding spaghetti on a spatula to an angry flood of water coming out of a pipe? I'm not sure a human drew this. My eyes aren't great lol there's a lol going on. Shelves are falling off the walls. Oh, actually, there are 2 other people, fleshing out the 4 kinds of thinkers described in the headline. But only the tops of their heads are visible so who knows what their problems are? I'm certainly not reading the story to find out. There are also some plants in there
July 25, 2026 at 7:29 PM
2bblue.bsky.social
Cluelessness, Shameless, Gutless, Thinkless, Meaningless, …
atrupar.com Aaron Rupar @atrupar.com · Dec 28
Trump: "I saw a very interesting President Putin today. He wants to see it happen. He told me very strongly. I believe him. Don't forget, we went through the Russia Russia Russia hoax together and he'd call me, I'd say 'Can you believe the stuff they're making up?' It turned out we were right."
December 29, 2025 at 5:12 AM