Saved
28 briefs, 66 sources read. Each bookmark is searched, quoted and graded, not summarised.
20 Oct
Alejandro AO compressed a 30-minute lecture into a 7-minute video using Opus 5.5 for scripting and Qwen3 TTS for narration. Worth trying: test whether Qwen3 TTS alone handles lecture-length narration acceptably before pairing it with a model.
19 Oct
Alejandro AO promises a tutorial on pi durable with checkpoints and crash recovery, but the linked page only defines the number 3.14159. The llama.cpp demo and the math constant don't match, so the tutorial content is unverified until the correct link is…
13 Oct
Edo Lunardi put together mkdir.directory, a free list of 34 web design inspiration directories. Worth bookmarking it before the next project starts, so I stop wasting time Googling "best design sites" at 11pm.
11 Oct
Diana and David put together a Google Doc of NYC food picks for 2026 World Championships attendees. Worth checking whether the list covers the neighborhoods you'll actually be in, since a masterlist can be too broad to use.
10 Oct
A scalper claims a 90.89% hit rate on MNQ 5-minute ORB extensions to the 50% level, while Lauren asks followers what they built with @bot this week. These two posts have nothing to do with each other.
9 Oct
Three posts from two accounts all make trading-performance claims with no verifiable numbers or sources attached. Worth checking whether either account has ever posted a real trade log or P&L screenshot before taking the next claim at face value.
7 Oct
One post pitches a four-month Codex App workflow with no proof beyond OpenAI existing; the other reports a real fine-tuning result on Qwen3.5. They have nothing in common. I'd skip the first and read the Unsloth repo to see if the 74.3 percent number holds up.
6 Oct
Two of the three posts are just screenshots of a trending tab with no analysis behind them. The Meta Rebalancer post is the only one with substance, but it links to nothing useful yet, so I'd wait for actual documentation before spending time on it.
5 Oct
Cybersec published a vulnerability hunting methodology and publicly credited Blackfrost AI for open-sourcing their Cyber-Frost Harness. Worth doing: read the Cyber-Frost Harness repo to see whether the methodology matches what Blackfrost actually built.
4 Oct
A Prime Intellect job posting for RL and distributed training work lists two book recommendations, and Hesamation called that out. Worth doing: read the two books before applying, or at least skim them so you can speak to them in an interview.
3 Oct
Avi flags AI biotech economics and recursive AI as megatrends, but the evidence provided only backs up the recursive AI claim. Worth doing: ask Avi for the supporting material on AI biotech before treating it as a confirmed trend.
29 Sep
Codie Sanchez says Joe Hudson coaches Sam Altman and other big-name leaders, but the only proof she offers is a Wikipedia page about an R&B singer. Worth doing: ask her to name the actual coach or link a real bio before repeating the claim.
27 Sep
Stoic asks how to compare perpetual futures and spot prices but names no platforms or numbers. Worth doing: ask him which exchange he uses so the question gets concrete.
25 Sep
Kris Abdelmessih promotes his Moontower Meta archive but provides no actual evidence of its content, while Gappy shares a concrete Google Drive folder of ten quant finance summaries. The two posts have little in common beyond both being finance-related;…
24 Sep
These two posts amount to claims without evidence: one maps hedge fund strategies to beta factors with no return data shown, the other defines "family" instead of describing the trading internship. Neither gives me anything to actually work with.
16 Sep
Teortaxes plans to run DeepSeek Harness inside Dataelem's DSH Desktop client. That's one person's test plan, so there's nothing to connect it to yet. Worth doing: see if anyone at Dataelem acknowledges the test before it happens.
17 Jun
Damnang made his paid Substack articles on co-packaged optics free so more investors can read them. Worth doing: read the articles and note which CPO companies he flags, since he's signaling he wants wider eyes on this specific AI infrastructure angle.
13 Apr
Alex Izydorczyk says AI trading bots tested out of sample show no persistent edge over random results. Worth doing: treat any bot's backtest as unproven until it survives a real out-of-sample window.
20 Jan
SixSigmaCapital reposts an old stock-entry article that includes no verifiable performance data, so it amounts to recycled opinion with nothing to check. Worth doing: ignore the repost and look for the original article's actual backtest numbers before…