Studies have shown that code generation LLMs produce low-quality output, require high levels of oversight, and have been found to introduce more bugs into code, with security vulnerabilities being a major reason companies have not adopted or have become cold to adopting LLMs.

The C4 dataset widely used for training language models has problems with composition and translation; much of its non-English content has been automatically translated from sources like Japanese patents, creating poor-quality training data that practitioners need to understand.

The Bender Rule states that when evaluating language models, you should explicitly name what language(s) you are considering the model for, since most claims about 'language models' implicitly assume English and do not properly account for multilingual limitations.

There is a fundamental misunderstanding of science in claims that LLMs will do science, imagining that anything making progress in LLMs is the only science, when in fact science across multiple fields uses diverse mechanisms and is funded through NIH and NSF grants which are being gutted at the federal level.

AI hype consists of multiple elements including comparing machines to humans (a pattern dating to 1956 when Marvin Minsky and John McCarthy claimed they could build circuits mimicking human thinking) and claiming that wide swaths of human tasks can be automated perfectly with no pain points.

The media frames booster vs doomer as the full range of reasonable discourse, but the speakers sit outside this frame, not in the middle, laughing at both positions equally because they both rest on unrealistic consciousness assumptions.

Anthropomorphizing LLMs as minds with intentionality and mutual meaning-making capacity with humans is dangerous because it grants them properties they do not possess and obscures the actual harms they cause through labor displacement and institutional devaluation.

Sam Altman's rhetorical move of saying 'I'm a stochastic parrot and so are you' (riffing on Bender's 'stochastic parrots' paper) represents a reduction of humans to mechanical word prediction machines, erasing what it means to be human.

A Pew survey of US workers found that only about 17% of US workers had used AI tools in any capacity, with 1% using them intensely and 16% using them sometimes, contradicting the narrative of widespread AI adoption and automation.

The singularitarian and transhumanist ideologies are fundamentally authoritarian, promising that centralized powers (Mark Andreessen, Sam Altman) will provide limitless abundance through AI while people lack control over their livelihoods and futures.

Claims by figures like Bill Gates that nursing, doctoring, and teaching will be wholly within the purview of LLMs in about 10 years are patently false because they fundamentally misunderstand what doctors, nurses, and teachers do—which is building relationships, aiding people through daily life, and comforting people in need, not simply dispensing knowledge.

The primary hypeman and beneficiaries of AI hype are institutional actors including billionaire founders (Sam Altman, Dario Amodei), their venture capital funders (Andreessen Horowitz, Sequoia Capital, SoftBank), tech company executives (Satya Nadella, Sundar Pichai, Brett Smithson), and credulous journalists with cultural capital (Kevin Roose, Casey Newton).

The appeal of technoutopian futures like Ray Kurzweil's singularity concept lies in the psychological desire to have someone else (technology, AI) solve problems for us rather than in collective action; this reflects a deeper desire to cede control and seek salvation through technological means.

Physiognomist AI systems that attempt to detect internal states (emotions, criminality, personality) from facial appearance replicate historical pseudosciences of phrenology and physiognomy, which have clear racial and ability-based harm histories.

About a third of artists and graphic designers have reported losing work due to the introduction of generative AI image tools; workers being rehired to correct outputs of these tools are being paid less than they would for original creation.

Enshittification occurs when a product that starts useful becomes degraded as companies prioritize profit over user benefit; LLMs like ChatGPT might not be enshittified on arrival but could follow this pattern as companies make their use mandatory for workers while removing the ability to validate outputs.

The AI industry is displaying characteristics of a speculative bubble; Sequoia Capital executive David Khan stated that AI companies must generate $600 billion in revenue, but current reported revenues are only $10-20 billion despite enormous investments in hardware and labor.

Satya Nadella acknowledged that the promised 10% productivity growth from the industrial revolution's steam engine is not appearing in LLM productivity data, suggesting that despite hype about AI driving enterprise productivity, the numbers do not support these claims.

Companies are throwing massive amounts of money at AI ('dying throws' of capital infusion like SoftBank's $40 billion investment in OpenAI) in a last-ditch effort to justify the bubble; this is a sign of desperation, not confidence.

When journalists like Kevin Ruse write credulous articles about AI rights and whether robots feel pain, they are doing bad journalism because they are not holding power to account, understanding the incentive structure behind AI narratives, or examining who benefits and who gets harmed.