1,691 source-backed ai facts - each attributed to a named person and a number, linked to the original source. Page 17 of 17.
“support for 112 languages”
“A 128GB gaming PC or Mac Studio can now run coding agents that build full stack applications.”
“last 18 months”
“A study by McKinsey found that companies using AI for automation saw productivity gains of up to 30%.”
“amassing over 200 million users by October 2024”
“The paper has been accepted as a Spotlight presentation at the top-tier machine learning conference, ICML 2025, and the code and dataset are now fully open-source.”
“model: 2012, unit: 44”
“In the pre-change scene the colourful insect had two stripes - one on each wing, and on the post-change, there was just one.”
“With the 7-day free trial on each plan”
“In 2026, a silent 128GB Mac Studio or a modern gaming PC runs coding models that rival frontier systems.”
“BCG research has shown that properly designed agentic AI systems can accelerate business processes by 30-50% and reduce low-value work by up to 40%.”
“a $20/month Plus plan, and a $200/month Pro tier”
“The dataset includes a wide range of failure logs collected from 127 LLM Multi-Agent systems, which were either algorithmically generated or hand-crafted by experts to ensure realism and diversity.”
“an iceberg scene with five penguins on it”
“Plaud AI is an transcription and note-taking service powered by the latest AI features that first hit the market in 2021.”
“DeepSeek broke the cost curve in early 2025.”
“Gartner recently estimated that out of the thousands of vendors claiming to offer agentic AI capabilities, fewer than 130 are genuinely doing so.”
“Launched in 2016”
“Even the best-performing single method achieved an accuracy of only about 53.5% in identifying the responsible agent and a mere 14.2% in pinpointing the exact error step.”
“Friday 11 June”
“Qwen3-Coder-Next hit 70.6% on SWE-Bench Verified with just 3B active parameters.”
“Launched in February 2024 Gemini”
“achieving 99.9% localization accuracy in unseen home environments”
“Mr Blair will provide strategic advice regarding investments in environmentally friendly or helpful technologies.”
“DeepSeek released R1 and V3”
“Bard which was introduced in March 2023”
“reducing overall trajectory error to approximately 2%”
“DeepSeek trained V3 on roughly 2,048 NVIDIA H800s for an estimated $6 million, versus the 20,000 A100s used for GPT-4.”
“Officially launched in November 2023 Grok”
“The paper has been accepted as a Spotlight presentation at the top-tier machine learning conference, ICML 2025”
“The stock market wiped roughly a trillion dollars off AI adjacent equities in a single session.”
“Founded in 2024 in Hangzhou”
“The dataset includes a wide range of failure logs collected from 127 LLM Multi-Agent systems”
“An NVIDIA H100 has 80GB of VRAM and costs about $30,000.”
“first introduced in March 2023”
“Using the Who&When dataset, the paper designs and assesses three distinct methods for automated failure attribution”
“A Mac Studio with an M-series Ultra chip gives you up to 512GB of unified memory for a fraction of that, and the CPU, GPU, and Neural Engine all share it.”
“Claude 3 family debuting in March 2024”
“Experiments were conducted in two settings: one where the LLM knows the ground truth answer to the problem the Multi-Agent system is trying to solve (With Ground Truth) and one where it does not (Without Ground Truth).”
“A modern gaming tower with 128GB of fast DDR5 and a mid-range GPU does something similar for even less.”
“three specialised models”
“The paper also introduces two key training strategies to further improve long-context performance:”
“A 128GB rig draws around 140W under load.”
“DeepSeek-V3, trained on a cluster of 2048 NVIDIA H800 GPUs, serves as a compelling case study...”
“It will happily host a 70B parameter dense model or a sparse 80B MoE with room to spare.”
“Table 1 in the paper compares the per-token KV cache memory footprint of DeepSeek-V3, Qwen-2.5 72B, and LLaMA-3.1 405B. DeepSeek-V3 achieves a remarkable reduction, requiring only 70 KB per token, significantly lower than LLaMA-3.1 405B’s …”
“Your local box is already busy running a 32B or 70B coding model at full memory bandwidth.”
“DeepSeek has pioneered the use of FP8 mixed-precision training for a large-scale MoE model. Despite NVIDIA’s Transformer Engine supporting FP8, DeepSeek-V3 marks a significant step as the first publicly known large model to leverage FP8 fo…”
“McKinsey still projects $5.2 trillion in data centre investment by 2030.”
“During EP parallelism, tokens are scheduled using fine-grained FP8 quantization, reducing communication volume by 50% compared to BF16, thereby significantly shortening communication time.”
“DeepSeek trained V3 on roughly 2,048 NVIDIA H800s for an estimated $6 million”
“DeepSeek currently utilizes the NVIDIA H800 GPU SXM architecture (Figure 2), which, while based on the Hopper architecture similar to the H100, features reduced FP64 compute performance and NVLink bandwidth (400 GB/s down from 900 GB/s in …”
“versus the 20,000 A100s used for GPT-4.”
“DeepSeek-V3, trained on a cluster of 2048 NVIDIA H800 GPUs, serves as a compelling case study demonstrating how a synergistic approach between model design and hardware considerations can overcome these limitations.”
“A Mac Studio with an M-series Ultra chip gives you up to 512GB of unified memory”
“DeepSeek-Prover-V2–671B, a model boasting 671 billion parameters.”
“It runs comfortably on a 64GB MacBook, never mind a 128GB Studio.”
“reached an impressive 88.9% pass ratio on the MiniF2F-test”
“LLM Scout's analysis of 15,252 AI queries and 90,232 extracted citations across ChatGPT, Claude, Gemini, and Perplexity between September 2025 and January 2026 shows a clear pattern:”
“successfully solved 49 out of 658 problems from PutnamBench”
“ChatGPT has always been link-light, averaging just over three citations per prompt.”
“ProverBench, a new benchmark dataset comprising 325 problems”
“Claude lost nearly two-thirds of its links per answer in three months.”
“ProverBench includes 15 problems formalized from recent AIME (American Invitational Mathematics Examination) competitions (AIME 24 and 25)”
“Gemini and Perplexity both lost roughly half.”
“The remaining 310 problems are drawn from curated textbook examples and educational tutorials”
“Zero‑citation answers are on average 56% longer than answers that include sources.”
“DeepSeek AI is releasing DeepSeek-Prover-V2 in two model sizes to cater to different computational resources: a 7B parameter model and the larger 671B parameter model.”
“Between October and January, citation counts declined slightly, yet average response length increased by 12%.”
“features an extended context length of up to 32K tokens”
“ChatGPT answers rebounded sharply in January to 770 characters, exceeding the October baseline.”
“DeepSeek-Prover-V2–7B, a smaller 7B parameter model.”
“In December to 477 characters,”
“88.9% pass ratio on the MiniF2F-test”
“Perplexity: In October, it delivered the longest answers of any model at 3,299 characters, signalling a premium, research‑grade experience. By January, average length had dropped to 2,161 characters, a 34% reduction.”
“49 out of 658 problems from PutnamBench”
“Gemini compounds this problem by offering the shortest responses at every citation level. Average length fell from 1,141 characters in October to 935 in January, an 18% decline.”
“SRPO has achieved impressive results on the AIME24 (50) and LiveCodeBench (41.6) benchmarks”
“Claude attempted a partial compensation. Response length declined from 1,211 characters in October to 870 in December, before rebounding to 1,249 in January, slightly above its baseline.”
“SRPO achieves this level of performance with only one-tenth of the training steps required by R1-Zero.”
“The same level of AI usage now produces roughly half the outbound click potential.”
“nearly 50% of the sampled groups within a batch produced identical rewards.”
“This alone can drive referral declines of 40-60%”
“By making GRPO 10 times more efficient, we can accelerate the development of AI models and make them more accessible to a wider range of applications.”
“Dataset: 15,252 AI queries, 90,232 citations”
“A 10x improvement in training efficiency could make reinforcement learning from human feedback accessible to much smaller teams and organizations.”
“Models analysed: ChatGPT, Claude, Gemini, Perplexity”
“April 15, 2025”
“up to eight times faster than DeepSeek‑R1”
“200 tokens per second on consumer‑grade GPUs – a staggering 50 times faster than human reading speed”
“smaller 9B parameter versions of both GLM‑4 and GLM‑Z1 models”
