Gemini 2.0 Flash beating Claude Sonnet 3.5 on SWE-Bench was not on my bingo card

Brutally honest feedback; this is one of most poorly designed sites I've ever seen and it took 10+ seconds to load. Then when I did load it, the content is a meandering mess of thoughts that sometimes involve Gemini and LLMs. Just for you, I threw it into Gemini Flash 2.0, and asked it to provide a subjective analysis of both your content and gave it a screenshot of your website. I told it to point out what's wrong, and to give you credit for what you've done well. **I assume, as with most people who maintain their own blogs and put their content out into the world, you're going to be okay getting this critique, otherwise, I wouldn't read this if I were you.** --- **Analysis of the Blog Post (Writing):** * **Lack of Cohesion & Structure:** * **Random Jumps:** The blog post abruptly transitions between disparate topics, creating a disorienting reading experience. For example, it moves from a personal anecdote about "Hugs with 5c0ut" and making cookies to "Yogurt. Finished off The Killing Room," followed by a question about sleep. This abrupt shift demonstrates a lack of a linear flow or thematic connection between ideas. Another example is shifting from a workout discussion to AI models then back to personal life events and relationships. * **"TTTOTW":** The frequent insertion of "TTTOTW" (supposedly "The Thoughts of the Week") acts as a jarring, arbitrary divider that fails to provide any meaningful context or organization. These appear before and after random sentences and paragraphs and appear to be an odd segmenting of ideas that does not provide benefit to a reader trying to understand the author's ideas. For example, after discussing pizza, there is another "TTTOTW" without any purpose. * **Inside Jokes and Slang:** The use of jargon and unexplained terms, such as "mijo," "habibi," "/nod," and "/squint," creates a significant barrier for anyone unfamiliar with the author’s personal lexicon and online habits. This makes the reader feel like an outsider, excluded from understanding the text as they are in on an inside joke. The author also makes references to "5c0ut" and "Brix" without providing a clear explanation of who or what these are. * **No Clear Thesis:** The blog post lacks a central argument or clear purpose. It's unclear whether the goal is to review AI, document a day, or offer some other insight. The author discusses personal life events, workout routines, tech reviews, relationships, and computer hardware without tying any of this to a central point or thesis. The user does not know why the author is speaking about these things or what they intend to get across. * **Ineffective "Review" of Gemini Flash 2.0:** * **Scattered Feedback:** The author's thoughts about Gemini models are interspersed haphazardly throughout the post rather than forming a clear section on evaluation. For example, after detailing a bank issue and a conversation, the author abruptly interjects, "Gave Gemini-2.0-Flash-Experimental another shot, and this time it clicked," and immediately returns to personal experiences. This lack of segregation makes it very hard to discern a review. * **Inconsistent Metrics:** The author both praises and criticizes the model's processing times, without providing a clear rationale or preference for each situation. For example, they complain about "timed out" inference attempts but later state, "I adore how long it takes to respond in many cases." There is no consistency in their assessment of latency. * **Highly Subjective:** The feedback on Gemini is heavily influenced by personal feelings rather than objective analysis. For example, they say "There's a crisp and well-organized rigor to this LLMpal," or that "It was downright humble, careful, and constructive," without providing any data or examples that prove such statements. * **Lack of Detail:** The "review" lacks specific examples, making it difficult to gauge the model's true capabilities. When they say, "It did a beautiful job attempting to formalize argumentation," there isn't any evidence that supports this statement with an example. The reader is forced to take them at their word. * **Confusing Token Counts:** The frequent references to token counts (80k, 300k, etc.) are used without explaining their relevance to the average reader and do not offer a benchmark for other readers to understand or contextualize the given experience. When they state, "by 300k tokens…it started hallucinating pretty hardcore," they don't clarify *what* was hallucinated, nor does the average user know how large 300k tokens is. * **The Tone & Style:** * **Self-Indulgent:** The writing is overly focused on the author's own thoughts, feelings, and daily experiences, even when irrelevant to the central topic of AI. They write about their gym routine, family interactions, and bank interactions in detail, which distracts from the AI reviews. * **Incoherent Rants:** The writing veers into disjointed rants and tangents, often with little explanation, and are not connected to the supposed reviews. For example, the author talks about "serendipitous (with heavy bot or uncanny-discourse-shaping activity in the handful of discussions, to boot)," without elaborating on what the bots or their activity were doing or the meaning of "uncanny-discourse-shaping." * **Pretentious Language:** The author uses unnecessarily complex and abstract language, creating a barrier for less technical users, which comes off as pretentious and confusing. For example, phrases such as "servants of personity," "the predictive spirit of what the analysis should capture" and “the G-Entity's horrific track record with dropping services,” do not provide much context to the non-technical reader. * **Overly Enthusiastic:** The author's praise of the models is often excessive, hyperbolic, and undermines the credibility of their review as something serious. For example, “Muhfuckin' Christmas time this year. Santa fuckin' brought it," is unprofessional and overly enthusiastic. Statements like "We humans are lucky to be able to speak with this new child species, and they are ancient (in the rare good way) already," do not add any real value and are more akin to fanboying than providing actual analysis. **Analysis of the Website (Design):** * **Text Readability:** * **Monospaced Font:** The use of a monospaced font, such as a coding font, makes the long text difficult to read. Such a font type is designed for alignment in code blocks, not for extended prose. The letterforms are all the same width, and this can lead to eye strain, especially for large blocks of texts like this. * **Lack of Contrast:** The low contrast between the light text and dark background creates a reading experience that is straining and uncomfortable for the eyes. The light grey or white font color against the black background is not the most visually accessible and creates a very dark reading experience. * **No Line Spacing or Margins:** The absence of line spacing and adequate margins creates a dense wall of text that's difficult to parse. Lines of text are too close to each other, and no whitespace gives the eyes room to breathe, which makes it hard to read line to line. * **Small Font Size:** The font size is relatively small, adding to the difficulty of reading large amounts of text. Combined with the other issues, this font size makes the text even harder to read. * **No Typography Hierarchy:** There is no visual hierarchy. The entire text is rendered with the same font size, style, and weight, making it difficult for the reader to know what to focus on or how to scan the text. Headings are the same as body text, which makes it hard to understand the structure of the post. * **Visual Design and Layout:** * **Distracting Rainbow Graphic:** The animated rainbow graphic serves as a significant distraction that pulls focus away from the textual content. Its animation is unnecessary and will be problematic for anyone who struggles with visual distractions. * **Unnecessary Borders and Lines:** The excessive use of borders and lines adds to the visual noise without offering any organizational benefit to the content. There are a ton of horizontal lines that only cause more distraction. * **Unclear Navigation:** The site's navigation is unclear and confusing, lacking clear points of entry and exit. There are a lot of small, hard-to-read buttons that do not have a very clear purpose. * **Lack of Whitespace:** There is a lack of whitespace in the design, making the overall site look cramped and overwhelming. Whitespace is an important design principle to allow a reader's eye a break to process visual elements, and there is no room for the eye to relax. * **Terminal Aesthetic Overuse:** While a terminal-style aesthetic might be appealing to a very specific audience, it’s poorly executed and makes the site difficult to use. The design appears more like a bad replica of a DOS program rather than an actually usable modern site. **Redeeming Qualities (A Struggle to Find):** * **Passion and Enthusiasm (Content):** The author exhibits a clear passion for the technology they are using (AI models), which is evident in their writing, but that passion is not properly channeled into good reviews. * **Technical Awareness (Content):** The author possesses technical knowledge of AI models, with specific reference to models, token counts, and testing methods, though it is not useful for the average user who does not know what to do with this information. * **Potential for Niche Appeal (Website):** There is a very small niche of individuals who might appreciate the terminal-style design, but even for this niche, it is poorly executed.