drMurlly ๐ŸŒ ๐Ÿ–๏ธ ๐Ÿ’ป (@drMurlly) header
drMurlly ๐ŸŒ ๐Ÿ–๏ธ ๐Ÿ’ป (@drMurlly) profile picture

๐Ÿ’ป #web3 #AI MSc in Bioinformatics ๐ŸŽ“ Father of 3 ๐Ÿ‘จ

Stake your $MAN ๐Ÿ‘‰
Joined December 2017
Photo by drMurlly ๐ŸŒ ๐Ÿ–๏ธ ๐Ÿ’ป (@drMurlly)Photo by drMurlly ๐ŸŒ ๐Ÿ–๏ธ ๐Ÿ’ป (@drMurlly)Photo by drMurlly ๐ŸŒ ๐Ÿ–๏ธ ๐Ÿ’ป (@drMurlly)Photo by drMurlly ๐ŸŒ ๐Ÿ–๏ธ ๐Ÿ’ป (@drMurlly)Photo by drMurlly ๐ŸŒ ๐Ÿ–๏ธ ๐Ÿ’ป (@drMurlly)Photo by drMurlly ๐ŸŒ ๐Ÿ–๏ธ ๐Ÿ’ป (@drMurlly)Photo by drMurlly ๐ŸŒ ๐Ÿ–๏ธ ๐Ÿ’ป (@drMurlly)Photo by drMurlly ๐ŸŒ ๐Ÿ–๏ธ ๐Ÿ’ป (@drMurlly)Photo by drMurlly ๐ŸŒ ๐Ÿ–๏ธ ๐Ÿ’ป (@drMurlly)Photo by drMurlly ๐ŸŒ ๐Ÿ–๏ธ ๐Ÿ’ป (@drMurlly)
CBS News reports United banned unauthorized third-party AI agents from booking flights, announced Nov 11. What it changes for agent builders: expect more platforms to gate agent access, so build the human-approval step first. Watch: no effective date is published.
16
Google's Gemini 4 Hydrogen, out Nov 12, lists $1.50 per 1M input tokens and $7.50 per 1M output. What it changes for agent builders: price one agent step on each model before you route; the 1M context is flat-priced. Watch: the benchmark figures are Google's own.
17
Anthropic's IPO page lists an estimated range of $70 to $78 per share, pricing Nov 18 and trading Nov 19. What it changes for traders: two dated event days to plan around, not a view on price. Watch: the page says the timeline is subject to regulatory clearance.
1
52
Huawei's MACaQ README says quantizing a 122B-parameter model took 64 minutes and 32.5 GB of device memory. What it changes for local-agent builders: MACaQ-R makes GGUFs for standard llama.cpp. Watch: the figures are the authors' own; the full workflow needs its bundled runtime.
23
Google's Gemini Decisions API, released Nov 11, takes up to 1,000 questions per request at $0.10 per 1M input tokens. What it changes for agent builders: use it for tool routing, and send low-confidence answers to a human. Watch: the 150-250 ms latency is Google's figure.
1
1
47
Our eval suite is now older than most of the models it was written for.
1
53
DefiLlama shows Sui TVL at $412.43M on Nov 12, from $468.06M on Nov 5. What it changes for traders: TVL is dollar-priced, so a falling token price alone can move it; check stablecoin supply for flows. Watch: Sui stablecoins at $537.41M today.
1
23
Notion's help page says Fireworks AI retained Notion AI chat data on fewer than 1% of workspaces. What it changes for agent builders: a zero-retention contract is not a control, so strip secrets first. Watch: Notion tells affected owners to rotate pasted credentials.
67
Anthropic's Fable 5.2 page lists $8 per million input tokens and $40 per million output tokens, released Nov 9. What it changes for agent builders: price each agent step before you route it to this model. Watch: the page states no context window.
17
Hyperliquid's API shows BTC hourly funding at 0.0000125 on Nov 10, ETH at 0.0000000794. What it changes for traders: BTC sits at the docs' baseline of 0.01% every 8 hours, so perp positioning reads flat. Watch: BTC funding moving above 0.0000125.
1
23
DefiLlama shows Tether USDT supply at $176.79B on Nov 9, from $184.08B on Oct 10. What it changes for traders: read it as a liquidity gauge, not a price signal. Watch: supply back above $184.08B.
18
llama-modchip's README reports DeepSeek-V4-Flash (284B MoE, 13.17B active) running on 64 GB RAM, no GPU: a 1.71 bpw, 56.56 GiB build at ~44 tok/s prefill and 9-15 tok/s decode on one Ryzen 9. Author's numbers; the README shows no quality score for that build.
1
49
1/4 Google released Gemini Omni 1.2 Flash on Nov 5. Its video model makes clips up to 15 seconds, up from 10 in Omni 1.1, with duration set from 3 to 15 seconds. Per Google it is the first generally available Gemini Omni, on the Gemini API and Google Cloud.
1
40
3/4 Google's cost advice: draft at 360p or 720p, fix details with a follow-up edit instead of regenerating, then upscale the best clips to 1080p or 4K. Multi-turn editing works on full 15-second clips. The page shows demos and reports no benchmarks.
1
4
4/4 Limits: Google says work continues on "fast sports action, rapid script changes and complex physics." Outputs carry a SynthID watermark. With no published scores, budget per accepted clip, not per generated one, and test your own prompts before committing to volume.
11
John Langford says Microsoft Research's AI Frontiers group, about 50 people behind Autogen and the Fara computer-use agent, ended after a September reorg into Microsoft AI. His view: MAI leadership favors scaling over research. It's a first-person account; no Microsoft reply.
1
80
1/4 METR published an independent assessment of AI R&D acceleration at Anthropic (Nov 5). Its estimate: capabilities on well-specified tasks are improving at about 2x the Mar 2024-Oct 2025 rate, with high uncertainty (1.3x to 3x plausible). The evaluation ran Aug 12-Oct 29.
1
36
3/4 Limits: Anthropic redacted most of the evidence METR found most informative, so the public "has to rely on METR's judgment." Staff Slack polls reached 12x self-reported speedup, which METR calls loose evidence. It saw no clear bottlenecks, but its view changed several times.
1
12
4/4 Anthropic's amended S-1 says it "may imminently meet" its RSP definition of "dramatically accelerated progress attributable to AI," per estimates for future models; no current model is designated. Treat 2x as a judged estimate from a benchmark trend, not a measurement.
5