AI Engineering Foundations: What Every Developer Needs Whats AI by Louis-François Bouchard 2026-09-25 | AI Engineering Foundations: What Every Developer Needs
Here’s how I research for my daily content Whats AI by Louis-François Bouchard 2025-08-28 | Here’s how I research daily AI videos and how KIVA can save hours per week 👇Normally: tabs everywhere → read releases, scan socials, cite-check, outline, publishWith KIVA: one click expands into related web searches, pulls social discussions, SEO ideas, outlines → doneTried it on my Gemma 270M news cover last week, workflow took minutes instead of 1.5 hours👉 Sounds cool? Try KIVA for research: bit.ly/kiva-louisThis video is sponsored, but I recommend trying out KIVA for free even without it.I’m Louis-François — PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀#ai #kiva #research
Why LLM evaluations are crucial Whats AI by Louis-François Bouchard 2025-08-27 | If you’re not measuring, you’re guessing. Here’s why LLM evaluations are the highest-ROI move 👇- Most failures come from bad specs, no real data, or models misapplying rules- Fix it with custom evaluations: JSON checks, tool errors, schema constraints, LLM-as-judge- Build a dataset, analyze real traces, and turn it into an "Analyze → Measure → Improve" loop- Best resource I’ve found: Hamel Husain & Shreya Shankar’s course👉 Grab my 35% discount link here: maven.com/parlance-labs/evals?promoCode=whatsai-louisThis video is sponsored, but this is genuinely the best course for evals out there.I’m Louis-François — PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀#ai #llm #evaluations #short
Gemini 2.5 Flash Image is insane Whats AI by Louis-François Bouchard 2025-08-27 | Gemini 2.5 Flash Image is just insane. Hands-down the best image editor I’ve tried in years 👇• Nails character consistency while changing outfits, poses, lighting, or scenes• Benchmarks show it topping image-edit leaderboards• “Thinking” model → understands intent, even fixes details you didn’t explicitly ask for (“make it look real”)• Supports multi-turn editing → refine iteratively (e.g. paint walls → then add furniture)• Creative power: merge up to 3 images, apply styles/textures, blend surreal compositions• Free in the Gemini app & AI Studio, or super cheap on API at $0.039/imageSure, small mistakes (text artifacts, cropped details) still happen — like every model. But the image editing focus is a huge leap forward. Definitely not just hype.I'll use that one for all my thumbnails for now on.Do you agree? Let me know your current favorite image generation model / app!I’m Louis-François — PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀#ai #gemini25 #imageediting #generativeai #google #imagen #texttoimage #aicommunity #short
OpenAI’s $200/Month Research Agent: Is It Worth It? Whats AI by Louis-François Bouchard 2025-08-26 | Transform the Way You Work: academy.towardsai.net/courses/ai-business-professionals?ref=1f9b29Master LLMs and Get Industry-ready Now: academy.towardsai.net/?ref=1f9b29Our ebook: academy.towardsai.net/courses/buildingllmsforproduction?ref=1f9b29In this video you’ll learn how Deep Research agents help:- Save hours of manual research- Generate structured, citation-backed reports- Analyze documents, PDFs, and data in context- Make smarter, faster, better-informed decisionsLearn more for free...Twitter: https://x.com/Whats_AISubstack (newsletter): louisbouchard.substack.com#chatgpt #deepresearch #llm
Is DeepMind’s Genie 3 the future of gaming ? Whats AI by Louis-François Bouchard 2025-08-25 | Is DeepMind’s Genie 3 the future of gaming ?• Generates controllable 3D environments from text prompts at 720p / 24 FPS• Worlds are interactive, consistent, and remember past actions (no re-gen where you already explored)• Runs only for a few minutes with ~1 min of visual memory• Biggest leap: long-horizon consistency in generated 3D spaces• But… can’t perfectly simulate real locations or handle multi-agent interactions → limits universality• More exciting for agents, simulations, and robotics research than for mainstream gaming (at least for now)Cool progress, overhyped as “gaming’s future”, but definitely a milestone in 3D generation.I’m Louis-François — PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀#ai #deepmind #genie3 #short
Mistral just dropped Codestral 25.08 Whats AI by Louis-François Bouchard 2025-08-22 | Mistral just dropped Codestral 25.08 — here’s my take 👇• Coding support in 80+ languages, 256K context• Comes with Devstral (agents), Codestral Embed (search), and Mistral Code (LLM) → full coding suite• Local-friendly + privacy-safe, good for enterprises• May lag behind giants like Gemini 2.5 out-of-the-box → real value needs customization or retraining, which adds cost• Best fit for mid/large companies that want to own their coding stackPromising, but not plug-and-play for everyone. Those are just my thoughts!I’m Louis-François — PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀#ai #mistral #codestral #llm #aicode #opensource #aicommunity #short
Is DeepSeek V3.1 a game-changer or hype? Whats AI by Louis-François Bouchard 2025-08-21 | Is DeepSeek V3.1 a game-changer, or mostly hype? Let’s break it down 👇What’s new:• Still a similar MoE transformer (671B params, 37B active) with 128K context• Merges DeepSeek V3 + R1 → one hybrid model with thinking (reasoning) and non-thinking (direct) modes — like what ChatGPT does with GPT-5• Major upgrades vs V3: – More efficient reasoning + stronger agent integration – Post-training boosted tool calling, substantial leaps on coding + tool-use benchmarks – Long-context quality got big investment: ~630B tokens @32K and ~209B @128K – Beats R1 in reasoning often times while answering fasterWhy it matters:• Open-source + pricing undercuts GPT-5 and others by 2–6x• “Thinking” mode enables deep reasoning chains, “non-thinking” mode keeps reliable JSON tool/function calling• Retrieval & summarization across large docs should degrade less thanks to long-context trainingHype check:• Probably the best open-source model right now, but benchmarks don’t always translate into real usage (I was underwhelmed whith DeepSeek V3 vs. closed alternatives)• Tool calling is split: – Strict JSON calls? Use non-thinking mode – Special search-agent format? That’s thinking mode onlyVerdict:Definitely not just hype. V3.1 is a strong, cost-efficient open-source contender. But like every model: you’ll need to test it in your own workflows before buying into the leaderboard glow...Do you use DeepSeek? What do you think? Drop it in the comments 👇I’m Louis-François, CTO & co-founder at Towards AI. Follow for tomorrow’s no-BS roundup.#ai #deepseek #deepseekv31
Google’s Imagen 4 is here, but is it worth your time? 🎨✨ Whats AI by Louis-François Bouchard 2025-08-21 | What’s new:• Family of 3 models: – Imagen 4 Fast → $0.02 per image, ~2.7s latency (10x faster than Imagen 3) – Imagen 4 → $0.04 per image – Imagen 4 Ultra → $0.06 per image, ~10s latency, higher detail (up to 2K res)• Max 20 API requests / min per project (higher limits possible on request)• Developers praise the speed, but note persistent text-rendering issues in complex imagesHype check• Fast model sometimes drifts from prompts — good for quick, simple generations• Standard & Ultra → better prompt alignment + quality, quite cheap for high-res outputs• Overall: speed + pricing make this a strong step forward in open access image generationWeakness is text on screen, especially vs. alternatives like OpenAI's.Let me know which news I should cover next, and I’ll tag you!I’m Louis-François, CTO & co-founder at Towards AI. Follow for tomorrow’s no-BS roundup.#ai #imagen4 #gemini #aistudio #texttoimage #generativeai #google #aicommunity
Gemini 2.5 Can Now Scrape Websites For You (URL Context) Whats AI by Louis-François Bouchard 2025-08-20 | Google just dropped a new game-changer for web scraping — here’s why it matters 👇What’s new:• Gemini 2.5 can now take links directly in prompts• Extract the whole page text or search for specific info• Works via API or Google’s AI Studio, just enable it• Handles up to 20 links per LLM call• Works on PDFs and images (but doesn’t actually “see” images or videos inside web pages. Be careful, it may hallucinate based on captions!)Why it’s interesting:• Next logical step for context engineering, LLMs pulling structured context in real time• More flexible than basic internet browsing: you can target specific links for your queries• With HTML tricks (asking it to give you image URLs, then calling it again), you can build a fairly complete scraperLimitations / hype check:• Doesn’t see images or video content on web pages directly• Won’t crawl multiple subpages automatically (you need to feed links manually)• Like any scraper, some sites crash it — debugging is hard since you don’t control what it “sees”. Also you sometimes need to say "yes you can do it" for it to understand it can indeed scrape HTML.• Costs tokens → if you send a long page, you pay moreVerdict: Not a replacement for full web scraping stacks, but a super accessible entry point for developers who just need quick, structured pulls of external dataDo you think this will replace scraping as we know it?I’m Louis-François, CTO & co-founder at Towards AI. Follow for tomorrow’s no-BS roundup.#gemini #llm #contextengineering
Is the Claude 4.1 update a game-changer or a small step? 🤔 Whats AI by Louis-François Bouchard 2025-08-19 | Let’s break it down 👇What’s new• Big boost in agentic tasks, coding, and reasoning → 74.5% on SWE-bench• Stronger at multi-file editing & precision (nice for coding!)• Opus 4.1’s coding focus feels like a practical shift toward real-world use cases (not just raw scale)Context• Might signal we’re closer to the pre-training plateau we’ve been talking about• Some devs even report practical superiority over GPT-5 for coding projects (Jeremy Howard called it a breath of fresh air vs. gpt-5)• Nice counterpoint for folks frustrated that ChatGPT forces GPT-5 by defaultHype check• Solid incremental update, not a revolution• But if you care about coding & agent workflows → this is a real upgradeWhich one do you prefer, Claude or ChatGPT?I’m Louis-François, CTO & co-founder at Towards AI. Follow for tomorrow’s no-BS roundup.#ai #claude #anthropic #claude41 #llm #coding #reasoning #agentic #swebench #gpt5 #aicommunity
Use LangExtract for data extraction ? Whats AI by Louis-François Bouchard 2025-08-18 | LangExtract might be your next preferred tool for data extraction.Let me explain…#llm #ai #data
Is Gemma 3 270m worth it? Whats AI by Louis-François Bouchard 2025-08-15 | Google just dropped what might be the best small LLM yet — should you use it?Let’s break it down 👇What we know:• Gemma 3 new model: 270M params• Pretrained + instruction-tuned releases• Great for text classification & data extraction (specific tasks) while cutting cost/energy• Strong instruction-following; 128K context like modern LLMs• Built on the Gemma 3 architecture so easier to fine-tune and use in existing pipelines• Cost-effective fine-tuningCaveats / hype check:• Out-of-the-box on complex tasks is limited → expect more hallucinations• Ground it with your data (RAG) and fine-tune for best results• Community notes: “dumb” responses without tuning (expected at this size)Hype or not?Not hype! Perfect for simpler tasks you fine-tune it on, and for fast, efficient local inference on smaller devices.Let me know which news I should cover next, and I’ll tag you!I’m Louis-François, CTO & co-founder at Towards AI. Follow for tomorrow’s no-BS roundup.#ai #gemma3 #llm
FLUX.1 Krea (dev) just dropped Whats AI by Louis-François Bouchard 2025-08-14 | FLUX.1 Krea (dev) just dropped with open-weights — is it overhyped?Let’s break it down 👇What we know:• SOTA (?) text→image with a focus on photoreal, “aesthetic photography” vibes• 12B rectified-flow transformer, directly distilled from [pro] → similar quality explains the strong results• Collaborative fine-tuning with RLHF• Claims to match closed solutions like their FLUX1.1 [pro] (not sure)• Single-GPU friendly; seconds to ~1 min per image• Variants on HF can speed things upNot hype / caveats:• Open weights democratize high-quality photo realism (building on SD & FLUX.1)• Non-commercial license limits “full openness”• Quality feels a bit overhyped vs Midjourney or OpenAI’s latest image modelsTry it, there are lots of free options online — tell me what you think!I’m Louis-François, CTO & co-founder at Towards AI. Follow for tomorrow’s no-BS roundup.#ai #generativeai #texttoimage
Is the new GLM-4.5 agent-focused model overhyped? Whats AI by Louis-François Bouchard 2025-08-13 | Is the new GLM-4.5 overhyped?Let’s break it down 👇What we know:• From a Chinese lab• MoE design: GLM-4.5 (≈355B total, 32 experts) & GLM-4.5-Air (≈106B, 12 experts) — similar to Kimi K2• Pushing open-source AI forward• SOTA open-model results; reportedly rivaling Claude Sonnet 4 on SWE-Bench (?)• Low API costs (~$0.20–$2.20 /M tokens) with integrations on Hugging Face, ModelScope, OpenRouter — priced like “mini” tiers• Trained with the Muon optimizer (again like Kimi K2) on high-quality code, synthetic tasks, long-context prompts• 128K context, hybrid “plan or respond” modes, and fast generation (up to ~100 tok/s)• Built-in agentic behaviorsMy take:• Feels very close to Kimi K2’s recipe — great for the community, but I’m not swapping my stack yet. Not an ideal size for local models.• If you’re already running open-weights at this scale, it’s an easy A/B to tryI’m Louis-François, CTO & co-founder at Towards AI. Follow for tomorrow’s no-BS roundup.#ai #glm45 #llm
Is Llama Nemotron Super 49B over-hyped? Whats AI by Louis-François Bouchard 2025-08-12 | Is Llama Nemotron Super 49B over-hyped?We’ll see… but here’s what actually ships 😉What we know:• New small reasoning model built from Llama-3.3-70B-Instruct• Multi-phase post-training: SFT (math/code/science/tool-calling) + RPO/DPO/RLVR for stronger chat, reasoning, and agentic skills• RLVR + NAS squeeze SOTA-level performance out of a mid-size model (less chasing giant parameter counts)• Among the top of the 70B class and even outperforms some larger models• Efficient: single H100 GPU, 128k context• Fully open + commercial: NVIDIA Open Model License, synthetic (Nemotron-Post-Training-Dataset-v1) + human (HelpSteer3) data, reward models, and RL toolkitCaveats / hype check:• 2023 data cutoff → pair with retrieval for freshness• Leaderboard wins may be overstated if you exclude top closed models (e.g., Claude 4.1, gpt-5)• Heavy synthetic data: TBD on generalization/bias trade-offsWhy it matters:• Transparent datasets + tooling → reproducible research• One-GPU footprint = accessible for labs & smaller companies• A “small thinking” model that punches above its weight• Totally open and reproducibleWho should try it?• If you already host a Llama-70B-ish model, it’s an easy A/B swap to test.I’m Louis-François, CTO & co-founder at Towards AI. Follow for tomorrow’s no-BS roundup.#ai #llm #opensource
Is Qwen-Image Worth it? Whats AI by Louis-François Bouchard 2025-08-11 | 🖼 Qwen-Image — hype or helpful?Qwen-Image is a promising new image generation model. 👇• 20B-param Multimodal Diffusion Transformer + 7B Qwen-2.5-VL encoder• Alibaba claims SOTA text rendering + strong editing vs. GPT & Flux• Trained on billions of curated image–text pairs (Nature 55 %, Design 27 %, People 13 %, Synthetic 5 %)• New MSRoPE (positional encoding) tech for better resolution scaling & image-text alignment• Apache-licensed (commercial use OK), FP8 weights lower fine-tune costsCaveats: heavy GPU needs for top quality, slower inference than smaller open models, “GPT-4o-level” claims likely overblown without broad benchmarks or user testing.Verdict: solid open-source progress in multimodal generation, but mind the hardware + hype gap.—Louis-François — PhD-dropout & CTO/co-founder @ Towards AIFollow for tomorrow’s no-BS roundup.#QwenImage #AlibabaAI #GenerativeAI
🎓 ChatGPT Study Mode... hype or not? Whats AI by Louis-François Bouchard 2025-08-09 | 🎓 ChatGPT Study Mode... hype or not?TL;DR: it's a prompt transforming ChatGPT into a tutor-style chatbot: questions, hints, self-reflection, scaffolded steps instead of a direct answer.• Personalization + quizzes/feedback to check understanding• Since it's (for now)_ just a smart prompt → behaviour can vary between users and interactions• Use it for active learning (think rather than getting answers), not cheatingVerdict: promising for study habits & 24/7 tutor access for improved learning. Excited to see where AI takes us regarding education.Will AI ruin or improve education? 👇—Louis-François — PhD-dropout & CTO/co-founder @ Towards AIFollow for tomorrow’s no-BS roundup.#ChatGPT #StudyMode #EdTech
GPT-5 vs GPT-5 mini vs GPT-5 nano Whats AI by Louis-François Bouchard 2025-08-08 | Stop overpaying!Don’t default to GPT-5. Here’s the quick pick guide 👇What they all share:• 400k context • images + text input, text output only • tool access• “Minimal reasoning” toggle = lower costUse GPT-5 when…• You need max reasoning/agent actions & quality over costUse GPT-5 Mini for…• Everyday dev + RAG (the 80/20 sweet spot)• ~5× cheaper/faster than GPT-5, higher rate limits• Earlier knowledge cutoff, fine if you rely on web/toolsUse GPT-5 Nano when…• Bulk summaries, labeling, quick replies at scale• ~25× cheaper than GPT-5, fastest outputsTL;DR: Quality = GPT-5. Value = Mini. Speed/scale = Nano.Save this & thank me later.— Louis-François, CTO & co-founder @ Towards AIFollow for tomorrow’s no-BS roundup.#GPT5 #OpenAI #LLM
GPT-5: hype or game-changer? Last fact is 🤯 Whats AI by Louis-François Bouchard 2025-08-07 | 🤖 GPT-5: hype or game-changer? Feels like chatting with a PhD-level expert? • 400 k context + “think-on-demand” → up to 80 % fewer thinking tokens (cheaper calls!) thanks to improved RL trainings. • Great with tools as most recent models. • Best OpenAI coding scores: tops SWE-Bench, 4o, & o3.• Multimodal (images + video in, text out) with stronger visual reasoning • Higher factual accuracy, calmer tone, fewer 🤯 emojis (finally!!!).• Knowledge cutoff Sep 30 ’24, but live internet access so not a big deal.• Free-tier access.• mini & nano variants too, sharing about which to use tomorrow, follow to know! Worth switching or overhyped? — Louis-François, PhD dropout & CTO/co-founder @ Towards AI. #GPT5 #OpenAI #GenerativeAI
gpt-oss-120b is useless Whats AI by Louis-François Bouchard 2025-08-07 | 🚀 gpt-oss-120b: hype or helpful?• 117 B-param MoE (5.1 B active), fits on a single 80 GB GPU• Near-o4-mini parity on reasoning, coding, math & health tasks• 128 k context, 3 “effort” levels, full CoT, tool use & code exec• Trained on English STEM/code, multilingual capacity probably quite bad• Skip it if energy costs, fine-tuning hurdles, or non-English matter to you; the 20 B may be a smarter first stop for efficiency, but still only trained for English.—Louis-François, PhD dropout & CTO @ Towards AI.Follow for tomorrow’s no-BS roundup.#OpenAI #gptoss120b #GenerativeAI
🧪 OpenAI’s gpt-oss-20b: hype or helpful? Whats AI by Louis-François Bouchard 2025-08-06 | • o3-mini-level results, but open-weight under Apache-2.0, so you can run it locally (and including for commercial purposes). • Mixture-of-Experts design → activates only a fraction of params per token; 128k context. • Runs on ~16 GB VRAM / laptop-class hardware; downloads on Hugging Face. • Tool use & agent-style workflows (browse/code/execute) are supported. • Caveats: strongest on text/STEM; multilingual & long-horizon reliability still TBD. Probably worse than other models since training was focused on English. Verdict: real progress in small-model efficiency + reasoning. Worth trying if you want local/edge deployments (and in english!), just don’t expect closed-model polish on day one.—Louis-François — PhD-dropout & CTO/co-founder @ Towards AI.Follow for tomorrow’s no-BS roundup.#OpenAI #GPTOSS #LLM #MoE #EdgeAI #AInews #HypeOrNo
Gemini 2.5 Deep Think: hype or helpful? Whats AI by Louis-François Bouchard 2025-08-05 | Gemini 2.5 Deep Think: hype or helpful?• New thinking model that runs parallel hypothesis streams and was re-trained (RL) to coordinate them, not just a prompt trick.• Results: SOTA on LiveCodeBench v6 and strong on Humanity’s Last Exam (vs models without tool use); also hit IMO (international math Olympiad) gold-medal standard in testing.• Access: Ultra-tier only in the Gemini app, around $140–$250/month.My take: useful for deep math/code/science, but for everyday tasks the compute + wait often outweighs gains. Try iterating a few times with standard Gemini 2.5 thinking (or o3) before paying Ultra prices.Will you try Deep Think? Did it make a big difference for you? 👇—Louis-François — PhD-dropout & CTO/co-founder @ Towards AI.Follow for tomorrow’s no-BS roundup.#Gemini #DeepThink #Reasoning
Can Anyone Really Code With AI Now? Whats AI by Louis-François Bouchard 2025-08-03 | The Vibe coding trend... hype or helpful?My take 👇• Vibe coding = intuitive, AI-assisted programming• Great for quick prototyping; doesn’t turn non-coders into pros• Best results with experienced devs + human verification (unit tests + reviews)• Not replacing senior engineers, likely replaces some intern-level tasks• Outcome: either teams ship more with the same headcount, or need fewer people to ship the same product. I bet most companies will choose to ship more work.What's your take on vibe coding?—Louis-François, PhD dropout & CTO/co-founder @ Towards AI.Follow for tomorrow’s no-BS roundup.#VibeCoding #ClaudeCode #llms
Claude Code: hype or helpful? Whats AI by Louis-François Bouchard 2025-08-02 | • Claude, but in your terminal—writes, runs, and refactors code• Real agentic workflows: plans steps, switches tools, executes• Why coding is a great agent use-case:1. the task is complex. beyond chat, you need to code2. high-value skill (look at software developer’s salaries)3. LLMs are strong on code, they’ve been trained on lots of it4. low cost-of-error → there are unit tests and human review before pushing to prod• Claude code is great for devs to ship more, but only if you learn what it’s doing• Rate limits & costs: usage can spike—self-reports show $$$; add spend caps, CI tests, and code review. Anthropic recently announced a rate limit for users.• Fuels the vibe-coding trend—more on that tomorrow! Follow to see it ;)Are you using Claude Code? What’s your verdict? 👇—Louis-François — PhD-dropout & CTO/co-founder @ Towards AI.Follow for tomorrow’s no-BS roundup.#ClaudeCode #Anthropic #AIagents
Figma Make is now open. Should you use it? Whats AI by Louis-François Bouchard 2025-08-01 | 🧩 Figma Make is now open. Should you use it?• Prompt→app: generate interactive prototypes & web apps from text (not just mockups).• Multimodal: attach designs/images in your prompt; it converts them into functional UI & code.• Great for PoCs and fast iterations, but for larger projects, plan on code review & security checks.• Real data: built-in Supabase integration lets you add auth, DB, and storage to your prototype.• Availability: now generally available (with plan-based limits).Verdict: promising step that democratizes app building. Hype for production, helpful for prototypes.Did you try Figma Make? Do you agree? 👇—Louis-François — PhD-dropout & CTO/co-founder @ Towards AI.Follow for tomorrow’s no-BS roundup.#FigmaMake #Figma #NoCode
Runway Aleph: hype or helpful? Whats AI by Louis-François Bouchard 2025-07-31 | 🎬 Runway Aleph: hype or helpful?• In-context video editing/generation with simple prompts (add/remove/transform).• Big step beyond Gen-3 Alpha: from text-to-video → in-context edits.• Demos show day→night, rain, reflection removal, green screen, scene extension & new angles.• Waitlist only so far—expect variability, artifacts on non-curated footage, and high compute costs (which means expensive to use).• Likely better identity/motion preservation vs older models, but real-world consistency is TBD.Verdict: Sora-level wow in demos; promising but early. Worth watching, especially for post-production.Are you joining the waitlist? 👇—Louis-François — PhD-dropout & CTO/co-founder @ Towards AI.Follow for tomorrow’s no-BS roundup.#Runway #Aleph #GenAI
Is Perplexity Comet Really Smarter Than ChatGPT? Whats AI by Louis-François Bouchard 2025-07-30 | ☄️ Perplexity Comet: hype or helpful?• Built on Perplexity’s AI search, it implements agentic behaviours like summarizing emails, organizing tabs, shopping/browsing, and auto-browsing with confirmations. • Early access, invite‑only; $200/mo (Perplexity Max). • Shift from passive web search to agentic web tasks; comparable to the recent ChatGPT agents. Power comes from Perplexity’s engine (Sonar + external LLMs). Hype: steep price, limited rollout, demo‑heavy right now; same privacy/safety concerns as any tool that can act for you.Value: if it genuinely saves you more than $200/month in time, power users may love it. Cool to see competition with OpenAI and other agent-focused big companies.Will you try it? 👇—Louis‑François — PhD‑dropout & CTO/co‑founder @ Towards AI. Follow for tomorrow’s no‑BS roundup.#Perplexity #Comet #AIagents
Conversational Segmentation (Gemini 2.5): hype or helpful? Whats AI by Louis-François Bouchard 2025-07-28 | 🖼️ Conversational Segmentation (Gemini 2.5): hype or helpful?• Segments with natural‑language prompts (vs Meta SAM’s points/clicks/boxes/masks).• Handles relationships, conditionals, OCR text, and multilingual labels.• Great for prototyping and LLM apps; edge/real‑time still favors YOLO/SAM.• Expect some run‑to‑run variability; not ideal for always‑on camera pipelines.• Best use: gemini‑2.5‑flash, avoid “thinking” mode, request structured JSON.Prompt to paste:Give the segmentation masks for the objects.Output a JSON list of segmentation masks where each entry contains the 2D bounding box in the key "box_2d", the segmentation mask in key "mask", and the text label in the key "label".Use descriptive labels.Hype or not? Tell me where you’d use this 👇—I'm Louis‑François, PhD dropout turned CTO & co‑founder @ Towards AI. Follow for tomorrow’s no‑BS roundup.#Gemini #GoogleAI #ComputerVision
Kimi K2 = Hype or Not? Whats AI by Louis-François Bouchard 2025-07-24 | Kimi K2 = Hype or Not?It is the first open‑source trillion‑parameter MoE model• 1 T total / 32 B active parameters → giant yet efficient• Outscores GPT‑4.1 & Claude‑4 (65.8 % SWE‑Bench, 53.7 % LiveCodeBench)• 128 k context, MIT‑licensed, no vision implemented… yet• Open weights, but self‑hosting needs mega GPUs, so most will hit the APIHype or next DeepSeek moment? 👇Good morning, I’m Louis-François, ex-PhD researcher turned AI builder at Towards AI. Follow for tomorrow’s no-BS roundup.#ai #llm #MoE
8 Big Changes Making AI Agents a Huge Deal! Whats AI by Louis-François Bouchard 2025-07-22 | Try Julius for free here: bit.ly/4lGSFTg (Use code WHATSAI30 to get 30% off for 3 months on the pro plan)►Full article and references: louisbouchard.ai/the-year-of-ai-agents► Our new book Building LLMs for Production: amzn.to/4bqYU9b►Twitter: twitter.com/Whats_AI►My Newsletter (My AI updates and news clearly explained): louisbouchard.substack.com►Join Our AI Discord: discord.gg/learnaitogetherHow to start in AI/ML - A Complete Guide:►louisbouchard.ai/learnaiBecome a member of the YouTube community, support my work and get a cool Discord role :youtube.com/channel/UCUzGQrN-lyyc0BWTYoJM_Sg/join#ai #agent #llms
Before You Build with AI, Watch THIS (Big Mistakes to Avoid) Whats AI by Louis-François Bouchard 2025-07-19 | Transform the Way You Work: academy.towardsai.net/courses/ai-business-professionals?ref=1f9b29Master LLMs and Get Industry-ready Now: academy.towardsai.net/?ref=1f9b29Our ebook: academy.towardsai.net/courses/buildingllmsforproduction?ref=1f9b29Learn more for free...Twitter: https://x.com/Whats_AISubstack (newsletter): louisbouchard.substack.com#ai #generativeai #llm
ChatGPT Hacks: 7 Tricks to Instantly Boost Your AI Results Whats AI by Louis-François Bouchard 2025-07-10 | Transform the Way You Work: academy.towardsai.net/courses/ai-business-professionals?ref=1f9b29Master LLMs and Get Industry-ready Now: academy.towardsai.net/?ref=1f9b29Our ebook: academy.towardsai.net/courses/buildingllmsforproduction?ref=1f9b29Learn more for free...Twitter: https://x.com/Whats_AISubstack (newsletter): louisbouchard.substack.com#ai #generativeai #llm
Making Prompts Robust and Ready for Re-Use - AI for Business Professionals Whats AI by Louis-François Bouchard 2025-07-07 | Transform the Way You Work: academy.towardsai.net/courses/ai-business-professionals?ref=1f9b29Master LLMs and Get Industry-ready Now: academy.towardsai.net/?ref=1f9b29Our ebook: academy.towardsai.net/courses/buildingllmsforproduction?ref=1f9b29Learn more for free...Twitter: https://x.com/Whats_AISubstack (newsletter): louisbouchard.substack.com#ai #generativeai #llm
Our Best Tips for Context Engineering (full training session) (from prompts to fine-tuning LLMs) Whats AI by Louis-François Bouchard 2025-07-04 | Link to the full course: academy.towardsai.net/courses/llm-primer?ref=1f9b29Personalization, databases, retrieval (RAG, CAG), frameworks, fine-tuning…We will discuss:LLM limitationsContext windowKnowledge issuesEmbeddings + encodersLong contextRAGData, vector databases, indexing, retrievalCAG (cached context)Fine-tuningRLFTIn five intensive sessions, learn how to go from a simple prompt all the way to production. Master the fundamentals, Retrieval-Augmented Generation (RAG), evaluation, agent workflows, and large-language-model optimization.Although there’s a “for developers” angle, the course is relevant to everyone—no prior experience required. A bit of Python knowledge helps but isn’t essential.Learn more for free...Twitter: https://x.com/Whats_AISubstack (newsletter): louisbouchard.substack.com#generativeai #rag #LLM
RAG vs CAG - A hands-on technical breakdown to choosing the right approach Whats AI by Louis-François Bouchard 2025-07-03 | Try AI Video Cut now and level-up your content workflow: aivideocut.com/?utm_source=WhatsAI&utm_medium=video&utm_campaign=youtube►Colab Notebook: colab.research.google.com/drive/1zBxl-7mU5rgl7V_zVXRfasHmFBFz7Tsv?usp=sharing►Full article and references: theneuralmaze.substack.com/p/rag-vs-cag-a-deep-technical-breakdown► Our new book Building LLMs for Production: amzn.to/4bqYU9b►Twitter: twitter.com/Whats_AI►My Newsletter (My AI updates and news clearly explained): louisbouchard.substack.com►Join Our AI Discord: discord.gg/learnaitogetherHow to start in AI/ML - A Complete Guide:►louisbouchard.ai/learnaiBecome a member of the YouTube community, support my work and get a cool Discord role :youtube.com/channel/UCUzGQrN-lyyc0BWTYoJM_Sg/join#llm #rag #cag
Automate half of your daily tasks with easy Python Scripts! Whats AI by Louis-François Bouchard 2025-07-02 | ...
MCP Explained: Universal API for Prompts, Data, and Tools Whats AI by Louis-François Bouchard 2025-06-28 | • MCP Auth SDKs with Descope: descope.com/blog/post/mcp-auth-sdk Full article and references: louisbouchard.ai/mcpMore links: • MCP 101: descope.com/learn/post/mcp • MCP explained in 15 mins (video): youtube.com/watch?v=GQDHxlKJe_M • Deepdive into MCP authorization: descope.com/blog/post/mcp-auth-spec • A2A 101: descope.com/learn/post/a2a • Enterprise Challenges in Deploying Remote MCP Servers: descope.com/blog/post/enterprise-mcp► Our new book Building LLMs for Production: amzn.to/4bqYU9b►Twitter: twitter.com/Whats_AI►My Newsletter (My AI updates and news clearly explained): louisbouchard.substack.com►Join Our AI Discord: discord.gg/learnaitogetherHow to start in AI/ML - A Complete Guide:►louisbouchard.ai/learnaiBecome a member of the YouTube community, support my work and get a cool Discord role :youtube.com/channel/UCUzGQrN-lyyc0BWTYoJM_Sg/join#agent #llm #mcp
AI Fundamentals Builders Need to Know to Fix LLM Limitations Whats AI by Louis-François Bouchard 2025-06-22 | Link to the full course: academy.towardsai.net/courses/llm-primer?ref=1f9b29In five intensive sessions, learn how to go from a simple prompt all the way to production. Master the fundamentals, Retrieval-Augmented Generation (RAG), evaluation, agent workflows, and large-language-model optimization.Although there’s a “for developers” focus, the course is still relevant to everyone—no prior knowledge required. A basic grasp of Python helps but isn’t essential to get value from the training.Learn more for free...Twitter: https://x.com/Whats_AISubstack (newsletter): louisbouchard.substack.com#generativeAI #AI #LLM
Reasoning Models Explained: The New Scaling Law Axis Whats AI by Louis-François Bouchard 2025-06-20 | Try out HeyBoss here with code "louisfrancois": https://heybossAI.com/?via=louisfrancois► Read the article version: louisbouchard.ai/reasoning-models► Join our upcoming Agents course (email or DM me to get in!): academy.towardsai.net/courses/agent-engineering?ref=1f9b29►Twitter: twitter.com/Whats_AI►My Newsletter (My AI updates and news clearly explained): louisbouchard.substack.com►Join Our AI Discord: discord.gg/learnaitogetherHow to start in AI/ML - A Complete Guide:►louisbouchard.ai/learnaiBecome a member of the YouTube community, support my work and get a cool Discord role :youtube.com/channel/UCUzGQrN-lyyc0BWTYoJM_Sg/join#llm #llms #reasoningmodels
LLMs Don’t Think—They Predict. Here’s Why That Matters Whats AI by Louis-François Bouchard 2025-06-16 | Transform the Way You Work: academy.towardsai.net/courses/ai-business-professionals?ref=1f9b29Master LLMs and Get Industry-ready Now: academy.towardsai.net/?ref=1f9b29Our ebook: academy.towardsai.net/courses/buildingllmsforproduction?ref=1f9b29Learn more for free...Twitter: https://x.com/Whats_AISubstack (newsletter): louisbouchard.substack.com#ai #generativeai #llm
A Brief Introduction to Large Language Models (LLMs) - AI for Business Professionals Whats AI by Louis-François Bouchard 2025-06-05 | Transform the Way You Work: academy.towardsai.net/courses/ai-business-professionals?ref=1f9b29Master LLMs and Get Industry-ready Now: academy.towardsai.net/?ref=1f9b29Our ebook: academy.towardsai.net/courses/buildingllmsforproduction?ref=1f9b29Learn more for free...Twitter: https://x.com/Whats_AISubstack (newsletter): louisbouchard.substack.com#ai #generativeai #llm
AI for Business Professionals - Towards AI Academy Whats AI by Louis-François Bouchard 2025-06-02 | Transform the Way You Work: academy.towardsai.net/courses/ai-business-professionals?ref=1f9b29Master LLMs and Get Industry-ready Now: academy.towardsai.net/?ref=1f9b29Our ebook: academy.towardsai.net/courses/buildingllmsforproduction?ref=1f9b29Learn more for free...Twitter: https://x.com/Whats_AISubstack (newsletter): louisbouchard.substack.com#ai #generativeai #llm
Python Fundamentals + CS Concepts — A One‑Stop Starter Class Whats AI by Louis-François Bouchard 2025-04-22 | Master the most in-demand skill for building AI-powered solutions—from scratch: academy.towardsai.net/courses/python-for-genai?ref=1f9b29Master LLMs and Get Industry-ready Now: academy.towardsai.net/?ref=1f9b29Our ebook: academy.towardsai.net/courses/buildingllmsforproduction?ref=1f9b29Learn more for free...Twitter: https://x.com/Whats_AISubstack (newsletter): louisbouchard.substack.com#ai #python #llm
AI for coding - Towards AIs Python Primer for Generative AI Whats AI by Louis-François Bouchard 2025-04-13 | Master the most in-demand skill for building AI-powered solutions—from scratch: academy.towardsai.net/courses/python-for-genai?ref=1f9b29Master LLMs and Get Industry-ready Now: academy.towardsai.net/?ref=1f9b29Our ebook: academy.towardsai.net/courses/buildingllmsforproduction?ref=1f9b29Learn more for free...Twitter: https://x.com/Whats_AISubstack (newsletter): louisbouchard.substack.com#ai #python #llm
The Future of Quantum from 14 CEOs! Insights from GTC 2025 Whats AI by Louis-François Bouchard 2025-04-08 | Deploy Your AI Agents 8x faster with LangWatch. Get a demo: langwatch.ai/?utm_source=louis-yt► Master the most in-demand skill for building AI-powered solutions—from scratch: academy.towardsai.net/courses/python-for-genai?ref=1f9b29► Master LLMs and Get Industry-ready Now: academy.towardsai.net/?ref=1f9b29►Twitter: twitter.com/Whats_AI►My Newsletter (My AI updates and news clearly explained): louisbouchard.substack.com►Join Our AI Discord: discord.gg/learnaitogetherBecome a member of the YouTube community, support my work and get a cool Discord role :youtube.com/channel/UCUzGQrN-lyyc0BWTYoJM_Sg/join#quantum #gtc #gtc25
