Resultados
23 artículosOrdenar
1:53EN
Matthew Berman· 8 jul 2026What Anthropic’s J-Space Research Suggests About AI ReasoningMatthew Berman · Forward FutureMatthew Berman discusses internal model representations, interventions and the question of consciousness, while the research itself describes important limits.
11:15ENMatthew Berman· 27 mar 2026ARC-AGI-3: Testing How AI Agents Learn the Rules of Unfamiliar GamesMatthew Berman · Forward FutureMatthew Berman explains ARC’s earlier grid puzzles and plays an interactive ARC-AGI-3 level, then examines early model results. The benchmark tests a specific form of adaptation rather than settling whether a system has general intelligence.
7:49ENMatthew Berman· 2 dic 2025DeepSeek V3.2: Reasoning, Sparse Attention and Agent TrainingMatthew Berman · Forward FutureMatthew Berman examines DeepSeek’s December 2025 model release, its benchmark results and the techniques used to improve long-context efficiency and tool use.
13:53ENMatthew Berman· 10 oct 2025Tiny Recursive Models and the Limits of Puzzle Benchmark ComparisonsMatthew Berman · Forward FutureMatthew Berman examines a small recursive network that reports strong ARC and Sudoku results, and what its experimental setup means.
15:13ENMatthew Berman· 10 may 2025Absolute Zero: Training a Reasoner on Self-Generated Coding TasksMatthew Berman · Forward FutureMatthew Berman examines a 2025 paper in which a pretrained language model proposes and solves code tasks using verifiable feedback.
14:04ENMatthew Berman· 29 abr 2025Qwen3: Open Weights, Adjustable Reasoning and Tool UseMatthew Berman · Forward FutureMatthew Berman reviews the April 2025 Qwen3 family, its benchmark claims, training pipeline and local demonstration.
19:06ENMatthew Berman· 25 abr 2025Sleep-Time Compute: Preparing AI Context Before a QuestionMatthew Berman · Forward FutureMatthew Berman reviews a 2025 paper on moving some reasoning ahead of a user query, and explains where its experiments found benefits and limits.
15:15ENMatthew Berman· 19 abr 2025Early Reactions to OpenAI o3 and o4-mini: Tool Use, Math and MistakesMatthew Berman · Forward FutureMatthew Berman reviews 2025 demos and benchmark reports, from location guessing and coding to simple failures that challenge the hype.
12:30ENMatthew Berman· 7 mar 2025QwQ-32B: A Smaller Reasoning Model and Its TradeoffsMatthew Berman · Forward FutureMatthew Berman examines Qwen’s reinforcement-learning recipe, tests a coding task, and compares benchmark scores and hardware requirements with DeepSeek-R1.
9:43ENMatthew Berman· 20 feb 2025Testing Grok 3 on Games, Reasoning, and Web SearchMatthew Berman · Forward FutureMatthew Berman tries a new set of audience-suggested prompts, from a Snake–Flappy Bird mashup to a family-logic puzzle and a tesseract visualization.
6:50ENMatthew Berman· 12 feb 2025DeepScaleR: A Small Reasoning Model Tested on Competition MathMatthew Berman · Forward FutureMatthew Berman reviews a 1.5B model’s reported benchmark results and runs a quantized version locally on one AIME problem.
8:52ENMatthew Berman· 21 ene 2025Industry Reactions to DeepSeek R1’s 2025 ReleaseMatthew Berman · Forward FutureMatthew Berman reviews early reactions to DeepSeek R1, from open-weight reasoning models and local hardware to benchmark results and API pricing.
21:02ENMatthew Berman· 24 dic 2024Reactions to OpenAI’s 2024 o3 Preview: Benchmarks, Cost and LimitsMatthew Berman · Forward FutureMatthew Berman surveys the responses to the o3 preview, from mathematical benchmarks to debates about AGI, compute and reliability.
24:08ENMatthew Berman· 12 dic 2024What Apollo’s 2024 Scheming Tests Found in Language ModelsMatthew Berman · Forward FutureMatthew Berman reviews experiments on deceptive actions, concealment and sandbagging, with the study’s staged goals and sandbox limitations in view.
26:43ENMatthew Berman· 10 oct 2024Testing OpenAI o1 on Planning, Rules and Spatial ReasoningMatthew Berman · Forward FutureMatthew Berman reviews a 2024 pilot study comparing o1-preview, o1-mini and GPT-4 on six planning domains.
10:57ENMatthew Berman· 13 sept 2024Testing OpenAI o1-preview: Tetris, Physical Reasoning and a North Pole ErrorMatthew Berman · Forward FutureIn September 2024, Matthew Berman runs the new reasoning-model preview through a personal rubric and examines where its answers succeed or fail.
21:21ENMatthew Berman· 12 sept 2024OpenAI o1’s 2024 Launch: Reasoning Benchmarks and Early TestsMatthew Berman · Forward FutureMatthew Berman reviews o1-preview and o1-mini, OpenAI’s reported evaluations, and his own early coding experiments.
12:03ENMatthew Berman· 6 sept 2024Reflection 70B: Self-Correction Claims and the Limits of a 2024 DemoMatthew Berman · Forward FutureMatthew Berman reviews a model trained to generate reflection steps. Its initial benchmark claims later proved difficult for independent evaluators to reproduce.
12:17ENMatthew Berman· 7 ago 2024Testing Gemini 1.5 Pro on Code, Reasoning, Images and VideoMatthew Berman · Forward FutureMatthew Berman compares standard and experimental Gemini versions in an informal 2024 review, finding mixed results and repeated output errors.
17:03ENMatthew Berman· 8 jun 2024Qwen2 0.5B and 72B: A Hands-On ComparisonMatthew Berman · Forward FutureMatthew Berman tests coding, arithmetic and reasoning prompts on the smallest and largest Qwen2 models in a 2024 demonstration.
14:04ENMatthew Berman· 15 dic 2023Mistral Medium in 2023: API Pricing and Informal Model TestsMatthew Berman · Forward FutureMatthew Berman compares the original Mistral Medium prototype with Mixtral and GPT-4, then tests coding, writing and reasoning through the API.
9:46ENMatthew Berman· 17 jul 2023Orca Mini V2 vs. OpenOrca: A 2023 Informal Model ComparisonMatthew Berman · Forward FutureMatthew Berman tests two Orca-inspired models on coding, writing and reasoning, with results shaped by prompt setup and subjective scoring.
9:58ENMatthew Berman· 26 abr 2023HuggingChat’s 2023 Debut: Testing Writing, Reasoning and Python CodeMatthew Berman · Forward FutureMatthew Berman tries the early OpenAssistant-powered HuggingChat on writing, arithmetic, logic and a Snake game, then compares a few outputs with GPT-3.5 and GPT-4. The results illustrate both promise and limitations in this small historical test.