Key Findings from Large Language Model Analysis

Most large language models are trained to predict the next token in a left-to-right (L2R) manner. However, Apple researchers discovered that right-to-left (R2L) models can significantly outperform L2R models on specific multiple-choice question (MCQ) tasks! I just read this new paper "Reversal Blessing: Thinking Backward May Outpace Thinking Forward in Multi-choice Questions" that challenges our assumptions about how language models process information. This "reverse thinking" approach uses Bayesian inference to evaluate answer choices based on their likelihood of generating the question, rather than the traditional approach of evaluating questions to predict answers. Surprising Results: The researchers trained both L2R and R2L models with identical data and computational resources across different model sizes (2B-8B parameters). - R2L models consistently outperformed L2R on logical reasoning tasks (LogiQA) - R2L excelled at commonsense understanding tasks (OpenbookQA, CommonsenseQA) - R2L showed dramatic improvement on truthfulness assessment (TruthfulQA - 51% better!) What's fascinating is that these improvements held across different model sizes, datasets, and random seeds, suggesting this isn't just statistical noise. Why Does This Work? The researchers explored three hypotheses for why R2L performs better on certain tasks: 1. Calibration - R2L naturally "auto-normalizes" different answer choices, avoiding the "surface competition" issue where semantically similar answers (like "dog" and "puppy") split probability mass 2. Computability - Different directional factorizations have varying computational complexity 3. Conditional Entropy - The optimal reasoning direction corresponds to lower conditional entropy Through controlled simulation studies with arithmetic tasks, they found strong evidence supporting the conditional entropy hypothesis - the direction with lower conditional entropy tends to perform better. Implications This research suggests exciting possibilities for future language model development: - We might benefit from models that can reason in multiple directions - Alternative factorizations beyond L2R and R2L could further enhance LLM capabilities - Task-specific reasoning directions could boost performance on targeted applications The study suggests that our default assumptions about "forward thinking" might not always be optimal.

Exciting breakthrough in LLM Research: A comprehensive survey reveals that Large Language Models (LLMs) are proving to be highly effective embedding models, marking a significant shift from traditional encoder-only models like BERT to decoder-only architectures. The research, led by scholars from Beihang University, University of Technology Sydney, and other prestigious institutions, demonstrates two primary approaches for deriving embeddings from LLMs: >> Direct Prompting Strategy • Leverages LLMs' instruction-following capabilities to generate topic-specific embeddings • Utilizes contextual representations for enhanced semantic understanding • Implements prompt engineering techniques for optimal embedding generation >> Data-Centric Tuning Approach • Employs supervised contrastive learning with carefully curated datasets • Incorporates multi-task learning frameworks for improved generalization • Utilizes knowledge distillation from cross-encoder models for enhanced performance >> Advanced Implementation Details The research reveals sophisticated techniques including: • Bidirectional contextualization for enhanced semantic capture • Low-rank adaptation for efficient parameter tuning • Integration of both dense and sparse embedding approaches • Implementation of innovative pooling strategies for token aggregation >> Performance Insights The study demonstrates remarkable improvements over traditional models: • Superior performance in classification, clustering, and retrieval tasks • Enhanced capability in handling long-context dependencies • Improved cross-lingual representation capabilities • Better scalability with model size and training data This groundbreaking research opens new possibilities for applications in information retrieval, natural language processing, and recommendation systems.

Large Language Diffusion Models (LLaDA) Proposes a diffusion-based approach that can match or beat leading autoregressive LLMs in many tasks. If true, this could open a new path for large-scale language modeling beyond autoregression. More on the paper: Questioning autoregressive dominance While almost all large language models (LLMs) use the next-token prediction paradigm, the authors propose that key capabilities (scalability, in-context learning, instruction-following) actually derive from general generative principles rather than strictly from autoregressive modeling. Masked diffusion + Transformers LLaDA is built on a masked diffusion framework that learns by progressively masking tokens and training a Transformer to recover the original text. This yields a non-autoregressive generative model—potentially addressing left-to-right constraints in standard LLMs. Strong scalability Trained on 2.3T tokens (8B parameters), LLaDA performs competitively with top LLaMA-based LLMs across math (GSM8K, MATH), code (HumanEval), and general benchmarks (MMLU). It demonstrates that the diffusion paradigm scales similarly well to autoregressive baselines. Breaks the “reversal curse” LLaDA shows balanced forward/backward reasoning, outperforming GPT-4 and other AR models on reversal tasks (e.g. reversing a poem line). Because diffusion does not enforce left-to-right generation, it is robust at backward completions. Multi-turn dialogue and instruction-following After supervised fine-tuning, LLaDA can carry on multi-turn conversations. It exhibits strong instruction adherence and fluency similar to chat-based AR LLMs—further evidence that advanced LLM traits do not necessarily rely on autoregression. https://lnkd.in/eYp9Hi5y

Language models don’t just predict words. They have actually figured out the mathematics of language. That's the mind-bending conclusion from a new paper from researchers of the paper “Language Models are Implicitly Continuous” that's challenging everything we thought we knew about how LLMs understand language . So what’s so special? We always assumed language models process text as discrete sequences of separate tokens - after all, that's how they're trained - to predict the next word in a sequence. But this fundamental assumption appears to be wrong. The researchers of this paper developed what they call a 𝗖𝗼𝗻𝘁𝗶𝗻𝘂𝗼𝘂𝘀 𝗖𝗮𝘂𝘀𝗮𝗹 𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺𝗲𝗿 (𝗖𝗖𝗧) extension that allows existing LLMs to accept continuous inputs without modifying their weights or architecture. This clever approach let them test how models respond to inputs that vary continuously in both time and space dimensions. 𝗞𝗲𝘆 𝗙𝗶𝗻𝗱𝗶𝗻𝗴 #𝟭 - 𝗧𝗶𝗺𝗲 𝗖𝗼𝗻𝘁𝗶𝗻𝘂𝗶𝘁𝘆: Imagine telling someone that when you say “apple”, you actually mean “half an apple”. Then two “apples” actually only means one. By shrinking token “durations” to a half, a third, or a quarter of their length, LLMs predicted this changed length without having to specify anything - ‘smaller’ tokens means smaller meaning of the word. The predictions from the LLM were 𝗰𝗼𝗻𝘁𝗶𝗻𝘂𝗼𝘂𝘀, 𝘵𝘩𝘦𝘺 𝘷𝘢𝘳𝘪𝘦𝘥 𝘴𝘮𝘰𝘰𝘵𝘩𝘭𝘺 𝘢𝘴 𝘵𝘰𝘬𝘦𝘯 𝘥𝘶𝘳𝘢𝘵𝘪𝘰𝘯 𝘤𝘩𝘢𝘯𝘨𝘦𝘥. This happened despite never being explicitly trained on token duration. 𝗞𝗲𝘆 𝗙𝗶𝗻𝗱𝗶𝗻𝗴 #𝟮 - 𝗦𝗽𝗮𝗰𝗲 𝗖𝗼𝗻𝘁𝗶𝗻𝘂𝗶𝘁𝘆: When researchers interpolated between embeddings of different words (like "apples" and "bananas"), the models treated these in-between states as semantically meaningful . For example, they might identify the colour of this hybrid concept as "green" - neither red (apple) nor yellow (banana). What makes this truly remarkable is that these models were never explicitly taught to handle continuous inputs. This behaviour emerges naturally from their architecture and training . 𝗪𝗵𝘆 𝗧𝗵𝗶𝘀 𝗠𝗮𝘁𝘁𝗲𝗿𝘀: - It challenges our fundamental understanding of how LLMs process language - It suggests they're learning representations that go beyond the discrete tokens they're trained on - It explains why LLMs sometimes perform well on rare or novel inputs - they're interpolating between familiar concepts - It reveals a profound difference between human and LLM language understanding - we think in discrete symbols, they operate *mathematically* in continuous space This wasn't just a one-model fluke. They observed the same continuous behaviors across Llama3, Llama2, Phi3, Gemma, Gemma2, and Mistral . So next time you're chatting with ChatGPT or Claude, remember - behind those discrete tokens is a mind that's secretly thinking in continuous space, understanding language in ways fundamentally different from our own. Read the full paper: https://lnkd.in/epZ4npgQ

🔍 New exclusive study powered by RESONEO: how people really interact with ChatGPT - Analysis of 87,725 public conversations I'm excited to share exclusive insights from a comprehensive analysis of nearly 90,000 ChatGPT conversations that users publicly shared. This data offers fascinating glimpses into how people engage with LLMs versus traditional search engines. Key Findings: 📊 Longer, More Complex Queries  - First messages average 8 words - more than double Google's 3.5-word average  - Users are asking more elaborate questions, not just keyword searches 💬 Extended Conversations Are the Norm  - 60% of conversations involve 5+ exchanges  - 35% extend beyond 8 messages This represents a fundamental shift from one-shot search queries to ongoing dialogue 🎯 Beyond Simple Fact-Finding  - Only 48.3% of queries seek factual information (where LLMs directly compete with search engines)  - 16.2% request opinions/suggestions  - 14.7% involve creative content generation This shows LLMs serve broader use cases than traditional search 💻 Technical Usage is Significant  - ~10% of conversations contain code  - GitHub ranks among top cited domains across all languages Developers are clearly embracing AI assistance 🌍 Global Patterns  - English dominates (55.5%), followed by Japanese (21.5%)  - Science & Tech topics lead (30%), followed by Politics & Society (18%)  The data reveals that LLMs aren't just replacing search engines - they're enabling entirely new interaction patterns and use cases. Important note: This analysis is based on conversations that users voluntarily made public through ChatGPT's sharing feature. While not representative of all ChatGPT usage, it provides valuable insights into how people interact with large language models.

𝗜 𝗵𝗮𝘃𝗲 𝗯𝗲𝗲𝗻 𝗶𝗻 𝘁𝗵𝗲 𝗡𝗟𝗣 𝘀𝗽𝗮𝗰𝗲 𝗳𝗼𝗿 𝗮𝗹𝗺𝗼𝘀𝘁 𝟭𝟬 𝘆𝗲𝗮𝗿𝘀 𝗻𝗼𝘄, and I know the first-hand challenges of building text-based models in the pre-GPT era! So, I am a 𝗽𝗿𝗼-𝗟𝗮𝗿𝗴𝗲 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗠𝗼𝗱𝗲𝗹 (𝗟𝗟𝗠) 𝗲𝗻𝘁𝗵𝘂𝘀𝗶𝗮𝘀t, but I don’t believe they will replace humans or solve all our problems, especially when it comes to highly complex reasoning in industries like Finance. 𝗧𝗵𝗶𝘀 𝘄𝗲𝗲𝗸𝗲𝗻𝗱, I spent reading two compelling papers, and I’m convinced we’re bumping into real reasoning ceilings: 𝗜> "𝗧𝗵𝗲 𝗜𝗹𝗹𝘂𝘀𝗶𝗼𝗻 𝗼𝗳 𝗧𝗵𝗶𝗻𝗸𝗶𝗻𝗴: 𝗨𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱𝗶𝗻𝗴 𝘁𝗵𝗲 𝗦𝘁𝗿𝗲𝗻𝗴𝘁𝗵𝘀 𝗮𝗻𝗱 𝗟𝗶𝗺𝗶𝘁𝗮𝘁𝗶𝗼𝗻𝘀 𝗼𝗳 𝗥𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝗠𝗼𝗱𝗲𝗹𝘀 𝘃𝗶𝗮 𝘁𝗵𝗲 𝗟𝗲𝗻𝘀 𝗼𝗳 𝗣𝗿𝗼𝗯𝗹𝗲𝗺 𝗖𝗼𝗺𝗽𝗹𝗲𝘅𝗶𝘁𝘆" (Apple) Apple researchers rigorously tested 𝗟𝗮𝗿𝗴𝗲 𝗥𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝗠𝗼𝗱𝗲𝗹𝘀 (𝗟𝗥𝗠𝘀), LLMs that explicitly generate chain-of-thought reasoning, using controlled puzzles like Tower of Hanoi and River Crossing Key insights: 1. 𝗧𝗵𝗿𝗲𝗲 𝗿𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝗿𝗲𝗴𝗶𝗺𝗲𝘀: ▪️Low complexity: standard LLMs outperform LRMs ▪️Medium complexity: LRMs excel ▪️High complexity: 𝗯𝗼𝘁𝗵 𝗰𝗼𝗹𝗹𝗮𝗽𝘀𝗲, accuracy plummets 2. Fascinating observation, 𝗟𝗥𝗠𝘀 “𝗴𝗶𝘃𝗲 𝘂𝗽” as puzzle complexity increases, their reasoning effort declines rapidly, even with enough tokens 3. Even when provided an exact algorithm (e.g., Tower of Hanoi strategy), the 𝗺𝗼𝗱𝗲𝗹𝘀 𝘀𝘁𝗶𝗹𝗹 𝗳𝗮𝗶𝗹𝗲𝗱 𝘁𝗼 𝗴𝗲𝗻𝗲𝗿𝗮𝗹𝗶𝘇𝗲 and mostly outputs based on past observed data pattern it is trained on 𝗜𝗜> "𝗥𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝗼𝗿 𝗢𝘃𝗲𝗿𝘁𝗵𝗶𝗻𝗸𝗶𝗻𝗴: 𝗘𝘃𝗮𝗹𝘂𝗮𝘁𝗶𝗻𝗴 𝗟𝗮𝗿𝗴𝗲 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗠𝗼𝗱𝗲𝗹𝘀 𝗼𝗻 𝗙𝗶𝗻𝗮𝗻𝗰𝗶𝗮𝗹 𝗦𝗲𝗻𝘁𝗶𝗺𝗲𝗻𝘁 𝗔𝗻𝗮𝗹𝘆𝘀𝗶𝘀" (Dimitris Vamvourellis & Dhagash Mehta, Ph.D., BlackRock) This study tested major 𝗟𝗟𝗠𝘀 (𝗚𝗣𝗧‐𝟰𝗼, 𝗚𝗣𝗧‐𝟰.𝟭, 𝗼𝟯‐𝗺𝗶𝗻𝗶, 𝗙𝗶𝗻𝗕𝗘𝗥𝗧 𝘃𝗮𝗿𝗶𝗮𝗻𝘁𝘀) on financial sentiment classification using: - "𝗦𝘆𝘀𝘁𝗲𝗺 𝟭" (𝗳𝗮𝘀𝘁/𝗶𝗻𝘁𝘂𝗶𝘁𝗶𝘃𝗲) - "𝗦𝘆𝘀𝘁𝗲𝗺𝟮" (𝘀𝗹𝗼𝘄/𝗱𝗲𝗹𝗶𝗯𝗲𝗿𝗮𝘁𝗲) 𝗽𝗿𝗼𝗺𝗽𝘁𝗶𝗻𝗴 Key takeaways: ▪️𝗥𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝗽𝗿𝗼𝗺𝗽𝘁𝘀 𝗱𝗶𝗱 𝗻𝗼𝘁 𝗶𝗺𝗽𝗿𝗼𝘃𝗲 𝗽𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲 ▪️Surprisingly, straightforward, intuitive prompts with GPT-4o (no chain-of-thought) outperformed all others  ▪️More reasoning led to overthinking, reducing alignment with human-labeled sentiments 💡 Why it matters for builders and researchers in Finance and every industry: ❎ 𝗕𝗶𝗴𝗴𝗲𝗿 𝗺𝗼𝗱𝗲𝗹𝘀 + 𝗺𝗼𝗿𝗲 “𝘁𝗵𝗶𝗻𝗸𝗶𝗻𝗴” = 𝗯𝗲𝘁𝘁𝗲𝗿 𝗼𝘂𝘁𝗰𝗼𝗺𝗲𝘀. Sometimes it’s actively worse ❎ We’re not seeing a soft plateau — these are 𝗵𝗮𝗿𝗱 𝗰𝗲𝗶𝗹𝗶𝗻𝗴𝘀 𝗶𝗻 𝗿𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝗰𝗮𝗽𝗮𝗰𝗶𝘁𝘆 ❎ For real-world systems, agents, and financial tools: design for 𝗿𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝗲𝗰𝗼𝗻𝗼𝗺𝘆, not just reasoning depth. #LLMs #ReasoningLimits #LLMChainofthought #LLMReasoningDecline

🔬 The Emerging Biology of Language Models I recently listened to the Latent Space Podcast with Emmanuel Ameisen and dived into the latest interpretability papers from Anthropic, and I think they represent a significant step forward in understanding what happens inside the AI black box. For a long time, many have viewed large language models as "stochastic parrots." This new research, however, provides compelling evidence that something much more complex and structured is going on under the hood. At the Englander Institute for Precision Medicine, we work to unravel the complex biology of human disease. I think it's fascinating to see a parallel approach emerging for AI. The researchers developed a method called "Circuit Tracing" which acts like a computational microscope. They build an interpretable "replacement model" that uses sparsely-active "features" instead of the model's hard-to-decipher neurons. By tracing the connections between these features in "attribution graphs," they can visualize the model's internal algorithms for specific tasks. The findings from applying this to Claude 3.5 Haiku are remarkable: 🧠 Internal Reasoning Models perform multi-step reasoning "in their head." To find the capital of the state containing Dallas, the model internally activates features for "Texas" before concluding "Austin". This isn't just memorization; the researchers showed they could swap in features for "California" and the model's output would change to "Sacramento". ✍️ Goal-Oriented Planning Models plan their outputs. When asked to write a rhyming poem, the model considers candidate rhyming words before it even starts writing the line. It then works backward from that planned word, constructing a sentence that leads to it naturally. 🌐 Abstract Generalization Models build language-agnostic representations of concepts. The same core circuits are used to identify antonyms in English, French, and Chinese, demonstrating a shared, universal "mental language". This reuse of circuitry is remarkable. For instance, the same pattern-matching circuit used for adding 36+59 is also activated to predict the end time of an astronomical measurement when it sees a start time ending in 6 and a duration ending in 9. 🕵️ Auditable Faithfulness We can begin to distinguish between genuine and unfaithful reasoning. The team showed instances where the model's written chain-of-thought was a fabrication, working backward from a hint provided in the prompt to derive an intermediate step, rather than computing it directly. I think the consequence of this work is a shift from treating models as inscrutable artifacts to seeing them as complex, yet scrutable, systems—an "in-silico biology" we can begin to map. This has profound implications for debugging, steering, and ensuring the safety of increasingly powerful AI systems. Podcast: https://lnkd.in/gABUvNpC Anthropic paper: https://lnkd.in/gYtWM2c4

🚨 Scaling laws are breaking, and not in a good way. Most ML teams still trust smooth scaling curves to guide training budgets and model deployment. But Nicholas Lourie, Michael Hu, and KyungHyun Cho find that only 39% of downstream tasks actually follow predictable scaling. The majority show inverse, noisy, or unpredictable behavior instead. 🔍 What they did: • Benchmarked 46 NLP tasks across model sizes, prompt formats, and validation sets • Compared scaling curves under different data and setup conditions 📉 What they found: • Many tasks show non-monotonic or erratic trends • Changing the validation set can flip which model performs “best” • Few-shot prompt tweaks can reshape curves entirely 📌 Why it matters: • You can’t blindly extrapolate from small models • Scaling laws aren’t universal. They’re setup dependent • Without internal curve plotting, you risk wasting compute chasing illusions 📎 Takeaway: Before scaling your next LLM, plot the curve for your actual task with your actual validation set. Don’t just trust the loss. And in sparse regimes, better annotation may beat bigger models. This work comes from researchers at New York University, Prescient Design, and CIFAR, and it’s a must-read if you still assume bigger always means better. Curious if others have run into “scaling curve surprises” in practice. Have you? #machinelearning #largelanguagemodels #scalinglaws #llmdevelopment #mlresearch

Can large language models really understand financial reports the way human analysts do? In finance, it’s not enough to produce a fluent answer — models must: - Extract precise values from dense tables - Chain multiple mathematical operations correctly - Show transparent reasoning (for audit & compliance) - And still compute numerically correct results - This gap between sounding right and being right is exactly what I explored in my latest blog, based on my research project in CMU’s Large Language Models course. 🔍 Key insight: Program accuracy ≠ Answer accuracy Even when an LLM generates the correct symbolic reasoning program, its own arithmetic often fails. External symbolic execution boosts numerical correctness from ~70% to ~99%, a strong case for neuro-symbolic systems in quantitative finance. ⚙️ What I studied: - In-context learning vs LoRA fine-tuning - Fine-tuning only 0.1% of model parameters (≈8M params, ~$3–4 training cost) - Achieving up to 9× improvement over few-shot prompting - Why ICL performance doesn’t predict fine-tuning potential - Where models still fail most (value extraction from tables) 📈 Takeaway: Parameter-efficient fine-tuning + symbolic tools offers a practical, cost-effective path to deploying LLMs for reliable financial reasoning — without retraining massive models. 👉 Full deep dive (methods, results, charts, code): 🔗 https://lnkd.in/eKXnwt6y Would love to hear thoughts from folks working on LLMs, finance, AI infrastructure, or applied ML. #LLMs #AI #Finance #MachineLearning #NLP #LoRA #Research #CMU #DataScience