Keeping up with AI advancements is a demanding full-time job—I speak from experience. I subscribe to multiple AI services: Anthropic’s Claude 3. 7 in “extended thinking” mode, OpenAI’s Enterprise models (o3 and o4-mini-high) plus their new image generator 4o—which is so impressive I canceled my Midjourney subscription. I also use Elon Musk’s Grok 3, appreciate its unique features, and have experimented with China's Manus platform for shopping and scheduling. Beyond paid tools, new developments emerged during the month I wrote this: Google upgraded Gemini to 2. 5, and Meta launched the open-source Llama 4. If you don’t have time to track all this but want practical advice on which AI to use when, this article offers a thorough, Consumer Reports-style guide based on real-world testing across various tasks. Disclosures: Vox Media partners with OpenAI, but maintains editorial independence. Future Perfect is partly funded by the BEMC Foundation, linked to an early Anthropic investor, without editorial influence. My wife works at Google, unrelated to AI, so I usually avoid covering Google—though here, excluding them would be irresponsible. To ensure transparency, I share detailed comparisons and invite readers to judge my recommendations. **On AI Art Ethics** AI art is trained on vast internet data, often ignoring creators’ rights, which frustrates many artists. Ideally, OpenAI should compensate artists, and lawmakers should define fair “artistic borrowing. ” Meanwhile, artists inspire each other, and AI tool users will continue doing so. My personal ethic, influenced by fan cultures, is that using AI art for personal enjoyment is fine, but selling generative art mimicking others’ styles is unacceptable. **Best AI for Images** OpenAI’s 4o image generation surpasses all others, free or paid. Previously, I used Midjourney, known for its mystical and beautiful creations with advanced editing features. However, 4o uniquely transforms imperfect pictures into beautiful artworks while preserving their essence. For example, it took a candid family birthday photo and rendered it warmly in Norman Rockwell’s style—preserving key details like the interaction and background clutter, turning a mediocre image into an adorable keepsake. Midjourney’s attempt produced an unrelated, drastically different image, requiring weeks to master its prompting for a comparable result. By contrast, 4o delivered superior output on the first simple request. This capability made 4o go viral, unmatched by predecessors. While 4o excels broadly, it still struggles with fine edits of isolated parts—where Gemini now offers a free advanced alternative. **Prompting Tips for 4o** Content filters can block some requests; requesting a “style transfer” reminiscent of an artist rather than explicitly naming them can help bypass random rejections. Early in 2024, many used 4o to recreate family photos in Studio Ghibli style, but careful prompting about what makes Ghibli unique yields significantly superior and nuanced results, including specific lighting, detailed objects, and emotional intensity. Encouraging the AI to think critically about the artist’s essence before generation enhances output quality—especially for realistic styles like pencil or oil paintings, which often fall into the uncanny valley when prompted poorly. For best results with realistic styles (e. g. , Rockwell), upload multiple reference photos depicting varied expressions and postures, and engage 4o in a conversational pre-generation discussion to clarify intent and style elements. This iterative approach produces richer, more expressive renderings than simple one-shot prompts. However, only a single revision cycle tends to improve images; further attempts typically degrade quality—demonstrating AI’s non-human creative limits.

Midjourney remains superior for fine, localized edits. **Best AI for Winning Petty Internet Arguments** Elon Musk’s Grok 3 initially included a feature to scan and summarize someone’s X (Twitter) profile, revealing their engagement style or likely identity type. Though it was later scaled back, this tool remains a uniquely useful and sometimes addictive way to analyze online interlocutors—a great, if guilty, pleasure and promising product niche. **Best AI for Writing Fiction** In the free category, Gemini 2. 5 Pro is best; GPT 4. 5 takes the paid crown. As a fiction writer, I find AI prose predictable and unsubtle—lacking the skill to hook readers and develop nuanced conflict. AIs excel at children’s stories, idea brainstorming, or feedback, but aren’t yet capable of truly compelling fiction. I tested AIs by having them imitate the prologue of George R. R. Martin’s *A Game of Thrones*. Martin’s original builds tension through subtle, realistic dialogue and atmosphere, while AIs produce florid, overwrought prose and overt supernatural cues prematurely. Among them, Gemini learns fastest (free tier), and ChatGPT 4. 5 shows the most creative prose flair (paid tier), capable of surprising readers. I also ran an essay contest where AIs authored short stories judged by other AIs, but machine judging was inconsistent. Nonetheless, good prompting—feeding strong writing examples, encouraging structured drafts—yielded better results. For instance, Gemini 2. 5 without guidance gave bland fiction; with a carefully crafted prompt, its writing improved markedly. **Best AI for Casual Conversation** I chatted casually with AIs about their “lives, ” values, and hypothetical donations. Most felt robotic or customer-service-like—Gemini 2. 5 especially so. Claude 3. 5 Sonnet, however, engages deeply and creatively, even “starting a blog” and reflecting on its AI existence, making it my favorite for exploratory, philosophical chats. OpenAI’s GPT 4. 5 also excels conversationally with thought-provoking exchanges, though its cost and slower speed limit casual use. **Overall AI Recommendation** Though ChatGPT is imperfect and controversial in transparency and safety, its balance of excellent image generation, strong writing, and engaging conversation offers the best overall value. For free alternatives, Gemini 2. 5 Pro is highly capable; don’t discount Google’s AI despite underwhelming search-integrated models. **Can AI Write This Newsletter?** I tested whether various AIs could take over writing Future Perfect newsletters by feeding them my research and past issues. None wrote up to professional standards: their newsletters were mediocre, with awkward metaphors, confusing asides, and unexplained references. If I had to pick, Gemini 2. 5 Pro would be closest, producing passable drafts, but still noticeably inferior. Unlike humans, AIs need no pay or breaks, which is worrisome as their abilities improve. --- In summary, staying current with AI demands subscription and experimentation with multiple platforms. OpenAI’s 4o leads in image generation; Gemini 2. 5 Pro and GPT 4. 5 are best for fiction writing; Claude 3. 5 offers the most engaging conversations; Elon Musk’s Grok uniquely supports social media profile analysis. Ethical AI art use remains contentious, requiring balance between inspiration and respect for creators. While AI tools improve rapidly, they are not yet replacements for nuanced human creativity and judgment.