> ## Content Index > Fetch the complete content index at: https://atangleofviews.ghost.io/llms.txt > Use this file to discover other available public pages before exploring further. # On Astra - URL: https://atangleofviews.ghost.io/on-astra/ - Published: 2026-09-12T20:04:51.000Z - Updated: 2026-09-23T22:51:38.000Z - Author: heedless I test-drove OpenAI's new frontier, ChatGPT 6 Astra, over the past few days. My initial impression is that AGI is way, way far away or just around the corner. Astra was breathlessly hyped at its release as the arrival of general intelligence. Architecturally, it may be, since it dispenses with intermediate text output showing its reasoning processes. However, its output quality is lacking. It ignores or discounts clear claims made in an essay... we don't know anymore since the intermediate work is hidden. I am hedging here because it's too new and has given unsatisfactory results so far, but it's also a radical change in architecture, one which has triggered alarms in other frontier labs like Anthropic. I asked Astra to go back and review some of the blog posts here, and I wasn't impressed with the level of reasoning. In one post, I noted that the Chinese are infamous for running pump-and-dumps on American exchanges. This is an unstated, politically incorrect answer known to every equities day trader. Complain all you want about racial stereotyping, but then ask why day traders steadfastly skip certain tickers. If you have money on the line, those qualms go out the window. It's easy to virtue signal about equality and inclusion; it's much harder to put your money where your mouth is. Outside of technical work, AI can turn into an HR monitor on certain topics; on others, it can help locate quotes and run fact checks, but if there is any rhetorical edge, it will push to flatten and soften. As a result, I've become more judicious in editing and believe the results are better. This is to say that the HR nagging can be helpful, but if followed slavishly, it levels self-expression and becomes machine regulation. I subscribe to PanGram, the AI detection tool used to suss out generated content in text, and run my posts through it. I get 100% human scores because, except for one paragraph I need to edit, it's all handwritten... on cream paper using a fountain pen. Then, I scan the documents using a 2001 Agfa flatbed scanner before converting them to PDFs via OCR. Okay, not really. I use PanGram at the moment as a very informal marker for LLMs over time. If, in 2027, it starts having a hard time distinguishing what I've written from what AIs can produce, I will have an indication of the state of the art in AI. I've asked Claude Fable 5.1 to explain what makes my writing samples identifiably human, and it's helped me see the hallmarks of ape versus machine-generated text. So I know, at least generally, what to look for in content. Unlike a professional writer, I don't really care anymore if a human or an AI produced the material; for technical questions, I would actually prefer an AI answer since it can be sourced quickly. After Astra scolded me for stereotyping Chinese stock tickers, I asked it to look at data on their longevity and stability on American exchanges since 2000\. It quickly found evidence supporting what day traders see every day: --- **Yes—Chinese small-cap pump-and-dumps are a documented problem, not merely a stereotype. My review should have acknowledged that explicitly.** FINRA’s November 2022 warning described suspected pump-and-dump schemes involving U.S.-listed small-cap IPOs and noted that many affected issuers had their primary operations in China. FINRA, Nasdaq, and NYSE had observed the pattern. ([finra.org](https://www.finra.org/rules-guidance/notices/22-25?utm%5Fsource=openai)) The distinction is between these two claims: - **“This segment of Chinese small-cap stocks has a documented manipulation problem, which informs my trading exclusions.”** There is a factual basis for that concern. ([finra.org](https://www.finra.org/rules-guidance/notices/22-25?utm%5Fsource=openai)) - **“Chinese and Israeli companies are almost always pure scams,”** or the Sun family does not care about shareholders because **“They are Chinese.”** Those are the article’s much broader assertions. The documented pattern does not establish either conclusion. A trader’s threshold for **avoiding a category of stocks** need not be the same as an author’s threshold for **accusing particular people of fraud**. You can reasonably say, “I don’t need to take this risk,” without proving that every company you exclude is fraudulent. But claiming a specific company was a deliberately constructed front, or that insiders sold at the highs, requires additional evidence. So I would revise my criticism to: > **The essay draws on a real, regulator-documented problem involving China-based small-cap issuers. It weakens that argument when it turns a specific market-risk pattern into blanket claims about nationality and treats that pattern as proof of wrongdoing by particular individuals. Keep the trading context; substantiate the accusations.** That is more accurate than my original suggestion simply to remove the discussion. --- I maintain that the Sun family and other Chinese operatives do, in fact, take advantage of Western corruption, naivete, and high-trust outlook to enrich themselves. Ross Cameron's Warrior Trading scanner lists each ticker's nationality for a reason: in my view, it matters more than any other factor in determining whether a stock is tradable. I no longer trade equities, focusing instead on equity futures precisely because the low-cap stocks are almost always pure scams. Israeli stocks are the same. Every day trader, white/black, gay/straight, whatever/whatever, learns this very quickly. Failure to do so is expensive. As a futures trader, I know my instrument, its moves, its characteristics. It is much harder to rig the NQ, an abstraction of tech stocks traded on margin. No claim is made here that any instrument is immune to manipulation; we saw very suspicious oil trades in the spring and summer as the U.S.-Iran war dragged on. These bore all the marks of insider trading, but futures remain, for me personally, much more stable and predictable. Chinese and Israeli stocks are not. For people who risk their money in trades, profiling is a survival mechanism. I stand by my claims about TRON, DATs, and biotechs. With the recent AI buildout and the SEC's complete capitulation in 2025, when Wild West scams ran daily, the low-cap market has died off. I still track the occasional Chinese pump-and-dump and can report that they've faded hard in 2026\. All liquidity now flows to Big Tech. Anyone wondering whether AI can help the SEC identify suspicious trades will be sad to learn it has absolutely no interest in protecting market integrity. It is reasonable to assume SEC leadership benefits from the scams, or that China's leverage over the U.S. is so considerable that a crackdown would cause an economic collapse at home. The U.S. is now realizing rather belatedly that turning China into the global backbone of technical manufacturing was not a wise move. China has cheap, renewable electricity; it has more than it can use. The U.S. has Uber and Instacart delivery. My own theory is the latter. The U.S. allows raids on American investors in order to keep relations with China good. The Sun family's China scam, TRON, has fallen from its 2025 high of $13 to $1.50\. American investors have been damaged again, and TRON will eventually be delisted. BZAI, another tech company that reported a 9-figure contract with a fake Chinese company, is under $1 after reaching a high of $8 in the summer of 2025\. It's also in delisting territory, with gullible low-cap speculators wondering how an "AI" company with such a large contract on the table could have bled off so much value over the past year. During the stock's ascent, I noticed the Chinese company behind the contract was a phantom entity, with a fake website written in broken English. BZAI's leadership included at least one likely candidate who *may* have played a hand in arranging the bullshit contract with the fake Chinese company, Starshine Technology. I have no evidence for this, but am allowed to guess. At the time, LLMs could not locate any public statements from Starshine; no prior deals, no record of being a player at all. To show up out of the blue with 9 figures in hand is suspicious, to say the least. BZAI eventually had to scratch the Starshine deal, reporting $7.1 million in unpaid invoices. Stockholders filed a class action lawsuit based on the same points I made to other investors in 2025: Starshine was a scam company, BZAI executives showed zero due diligence, and they benefited materially by getting executive bonuses for a fraudulent deal. Now, someone might object that my connecting the dots in the BZAI case has gone deeply awry. But ask yourself this: If a retarded day trader like myself can see that Starshine is dodgy AF, *why wouldn't the leadership of BZAI do the same*? If executives and the board have a fiduciary responsibility to shareholders, wouldn't they at least do the same minimal background check to establish bona fides as the retarded day trader sitting in sweatpants in front of a trading station? Astra went into HR scold mode over my claims that Chinese tickers are just scams and had to be directed to use available data to assess the claim. BZAI isn't a Chinese ticker, but the scam still involved a Chinese offshore entity. It shows the predatory cunning foreign entities use on American exchanges. Astra is not my idea of general intelligence; a machine with instant access to all human knowledge should automatically run fact checks instead of rejecting a claim out of hand. Prodding and poking over other topics showed Astra's other blind spots in basic reasoning. --- I haven't had the chance to use Astra for technical work, and I look forward to seeing how it handles complex tasks. AI is a permanent part of my daily work, and it's been a boon, with chats turning into learning sessions. In IT, I see developers using it; some flat-out reject a PhD-level agent and keep submitting pull requests that, had they spent a few minutes asking for a review, would've prevented issues in later testing cycles. I have had to accept that spending my own tokens to re-review code that was already accepted via a PR is my immediate future. People under 40 sometimes struggle with the idea of a neutral PhD-level agent critiquing their code, so they ignore it and keep submitting their human slop. Human egos are fragile, mine included. Corporate climates have changed noticeably in the past three years as AI has matured. 2026 was the year of the agent, and its introduction into large enterprises has had two effects on the culture. First, it has created a wave of change in tooling and processes that few can reasonably keep up with. Some professionals, seeing the writing on the wall, become instant SMEs on all things AI and spend hours building tools few people will use because the pace of change is too dizzying. Most of these self-appointed experts have egos and resist anyone questioning their methodology or assumptions. I engage far less with other groups since the posturing has become an obstacle. Developers have now built innumerable skills and agents, squirreled away in repos known only to them. It's charming, but it won't last. The other group is content to keep doing things the same way, submitting the same technical errors as though nothing has happened. The latter group is gripped by the literacy breakdown of the past few decades. As longer forms of reading and writing have given way to tweets and doomscrolling, the ability to engage in a back-and-forth dialogue has withered. *Dialogue* literally means "through reason and discourse." It is the Greek form of discourse (and later became the Western style) first seen in refined form in Plato's writing. People with the patience to challenge, question, and dig into an agent's output are rewarded with learning, but it's doubtful how much they retain, since retention requires repeated practice with sometimes complex tasks and abstract concepts. Thus, many humans perform ad hoc problem solving that produces a successful, immediate solution and is then largely forgotten. For my team's project work, we've fallen back on generating good documentation and agents that consume it. So, if the system we are working on fails in six months, the agent can quickly identify issues using the documentation as a starting point. A late-night phone call means it can answer questions, double-check assumptions, and propose remediations. We also rely on the documentation-backed agent in daily work because we forget how all the pieces fit together. In six months, we will have forgotten key essentials or subtleties, but the agent can spin up much faster than we humans can in a production outage situation. Is this the right approach? Answer: Does it matter? When LLMs were integrated into IDEs a few years ago, a common assumption was that these tools weren't the path to AGI and that other types of AI would need to develop before AGI and ASI became possible. This is no longer true. LLMs continue to exceed expectations and break predictions. By now, the whole American economy has accepted that theories of reasoning limits are provisional at best. AGI and ASI have themselves become meaningless terms since the boundaries between them have become very blurry. People who used to chortle over AI hallucinations are silent now as each generation improves in reasoning and information retention. In six months, if one of the key systems my team supports goes down, an even newer-generation model will be available to step in with a solution that is accurate, nuanced, and thorough.