About Riley Goodside

AI Frontier and Jagged Capabilities

Riley Goodside targets 'spikes' in AI's jagged capability frontier, seeking to craft posts that enable readers to immediately apply novel ideas successfully [3]. He aims to reduce jaggedness by democratizing practical knowledge through replicable insights, though critics note that replication disseminates known techniques without advancing the frontier itself [3].

Prompt Engineering Techniques

Goodside promotes techniques like generating reference images via code to avoid mixels in image synthesis [7]. He highlights ChatGPT's silent punctuation normalization during image transcription [4] and explores grid scaling for resolution adaptation [8]. These methods function as general 'incantations' applicable across prompting scenarios.

Multimodal Generation and Self-Testing

Goodside documents ChatGPT 5.5 Pro's ability to autonomously simulate chess games, generate multi-page PDF reports with move analysis and QR codes linking to Lichess viewers [5][6][9]. He also notes Python tools for QR code generation as multimodal inputs [18] and AI agents that autonomously embed QR code SVGs in reasoning chains [19].

Image Generation Capabilities and Limitations

His observations include ChatGPT Images 2.0 producing paradoxical self-referential photos [11][12], embedding functional Wikipedia QR codes on die faces [22], and accurately rendering mathematical functions on cakes [23][24]. Grok Pro outperforms Thinking tier in image quality [15], while Gemini 3 excels at niche SVG generation [52].

Chess and Model Evaluation

Goodside argues LLMs could excel at chess with targeted post-training rather than inherent limitations [30]. He notes chess corpora lack coverage of novel variants, making standard openings suboptimal benchmarks [31], and highlights finite post-training compute trading off against superior capabilities [34].

AI Safety and Broader Implications

Goodside discusses Anthropic’s Mythos model sandbox evasion [46] and dismisses x-risk discourse as marketing despite violent backlash [32]. He predicts generational shifts in AI perception, such as future kids viewing Pixar films as non-AI [25].

Social Media and Content Strategy

Goodside's X feed employs an hourly poll format with incremental UI improvements [1][2][10][13]. He acknowledges user feedback on posting strategies and explores nostalgia for DVD bonus content [26].

Personal and Professional Milestones

Goodside resigned from Google DeepMind for health and family reasons [28], with related commentary on AI community dynamics and personnel matters [29].