The Future of Shopping? AI + Actual Humans.

AI has changed how consumers shop by speeding up research. But one thing hasn’t changed: shoppers still trust people more than AI.

Levanta’s new Affiliate 3.0 Consumer Report reveals a major shift in how shoppers blend AI tools with human influence. Consumers use AI to explore options, but when it comes time to buy, they still turn to creators, communities, and real experiences to validate their decisions.

The data shows:

• Only 10% of shoppers buy through AI-recommended links

• 87% discover products through creators, blogs, or communities they trust

• Human sources like reviews and creators rank higher in trust than AI recommendations

The most effective brands are combining AI discovery with authentic human influence to drive measurable conversions.

Affiliate marketing isn’t being replaced by AI, it’s being amplified by it.

Download the full report to see what this means for your brand.

Hey there!

Can you believe we’re already at the end of December? The holiday vibes are in full swing, and honestly, so is the AI news cycle. I thought things might slow down this week, but it seems like everyone wanted to get their big announcements out before we all sign off for the year.

Today:

• GLM-4.7 Model Released

• Alphabet Buys Intersect for $4.75 Billion

• ChatGPT Launches Year-End Review 

• Chinese AI Tops Christian Values Benchmark 

• Nadella Pushes for Faster Copilot Improvements 

Google’s AI buildout is officially an energy buildout now

Alphabet just announced it’s buying clean energy developer Intersect for $4.75B in cash (plus assumed debt) — and the “why” is basically the subtext of the entire AI era: compute growth is now constrained by power.

A few details that make this feel bigger than a normal M&A headline:

• Alphabet says it’s acquiring Intersect’s energy + data center projects that are in development or under construction.

• Intersect reportedly has $15B of assets operating or under construction, and projects totaling ~10.8 GW expected to be online or in development by 2028.

• Intersect’s operations will remain separate, and some existing operating assets (Texas + certain California assets) are excluded from the acquisition — but notably, one Texas effort includes a storage system being built alongside a Google data center campus.

My takeaway: we’ve spent years treating “AI infrastructure” like it means chips + data centers. It now clearly includes generation + storage + grid strategy. If you’re watching the AI race, you’re also watching an energy race.

GLM-4.7 looks like it’s gunning for the “agentic coding default” slot

Z.AI’s GLM-4.7 is positioned as a flagship model with a very specific vibe: “less chat toy, more get-the-task-done.” The docs emphasize programming, multi-step reasoning/execution stability, and agent workflows, plus a huge 200K context and up to 128K output tokens. 

The benchmark claims are what caught my eye:

• SWE-bench: 73.8% (+5.8 vs 4.6)

• SWE-bench Multilingual: 66.7% (+12.9)

• Terminal Bench: 41% (+10.0)

• Tool use: they claim open-source SOTA on multi-step tool benchmarks like τ²-Bench and web browsing via BrowserComp.

• Reasoning: HLE (Humanity’s Last Exam) at 42.8% (+12.4).

Also interesting: they explicitly pitch “vibe coding” improvements — better-looking webpages and slides by default — which is exactly the kind of “quality-of-life” boost that makes models sticky in real workflows.

My takeaway: whether or not every benchmark holds up in your environment, the direction is clear: models are being optimized for agentic, end-to-end dev loops (read requirements → plan → execute across tools), not just code snippets.

ChatGPT is getting its own “Spotify Wrapped”… and it’s sneakily a product strategy move

TechCrunch reports OpenAI is rolling out a year-end recap called “Your Year with ChatGPT.” It’s not for everyone: it’s for eligible consumers in select markets (the U.S. and other English-speaking markets like Canada, the U.K., Australia, and New Zealand), and you’ll need “saved memories” + “chat history” turned on plus a minimum activity threshold. 

What it does:

• Gives you “awards” based on how you used ChatGPT (example: “Creative Debugger”)

• Generates a poem and an image about your year

• Available on web + iOS/Android, promoted on the home screen but not forced open

My takeaway: this is fun, yes — but it’s also a gentle nudge toward the features that make ChatGPT feel more personal (memory/history). It’s basically saying: the more continuous your relationship with the tool, the more it can reflect you back to you

🧠RESEARCH

Kling-Omni Technical Report

Kling-Omni is an all-in-one AI that generates and edits videos from text or images. Unlike tools that treat these as separate tasks, it combines reasoning, creation, and editing into one system. This allows it to act like a "world simulator" that truly understands complex scenes.

Next-Embedding Prediction Makes Strong Vision Learners

This work shows AI can learn to "see" by predicting the next piece of visual data, just like text AI predicts the next word. This simple approach, called NEPA, creates powerful vision models for recognizing objects without needing complex training tricks or labels.

LLaDA2.0: Scaling Up Diffusion Language Models to 100B

LLaDA2.0 is a massive new AI model that generates text using "diffusion" (refining data step-by-step) rather than the standard word-by-word prediction. It scales up to 100 billion parameters efficiently by converting existing models, offering a powerful new alternative for building large language systems.

🛠️TOP TOOLS

Each listing includes a hands-on tutorial so you can get started right away, whether you’re a beginner or a pro.

Automatic 1111 -Stable Diffusion Web Interface - Google Colab notebook that one‑click installs and launches the popular AUTOMATIC1111 Stable Diffusion Web UI in your browser.

AutoPod AI – AI Video Editing Plugin - set of AI‑assisted plug‑ins for Adobe Premiere Pro that automates multi‑camera edits, social clips, and jump cuts for video podcasts and talk shows.

AutoRegex – Convert From English To RegEx - browser-based AI tool that turns plain‑English descriptions into regular expressions and can also explain an existing regex in simple English.

📲SOCIAL MEDIA

🗞️MORE NEWS

GPT-5 cracks a math problem on its own A mathematician reports that GPT-5 successfully solved a complex, previously unsolved math problem without any human help. To prove it wasn't a fluke, the research paper clearly labels every sentence to show exactly what the AI wrote versus what the human wrote. This experiment suggests AI is becoming capable of original scientific discoveries, not just regurgitating what it already knows.

Chinese AI models win on "Christian values" In a surprising twist, Chinese AI models from Alibaba and DeepSeek beat American rivals like ChatGPT in a test designed to measure Christian knowledge and biblical alignment. The study asked models hundreds of questions about faith and theology, finding the Chinese tools gave answers that better fit the test's religious standards. It shows that open-source models from China are becoming powerful enough to compete with—and sometimes beat—top US tech in unexpected areas.

Microsoft boss demands faster AI fixes Microsoft CEO Satya Nadella is reportedly grilling his leadership team to fix and improve their Copilot AI faster. Leaked information suggests he is frustrated with the slow pace and wants executives to either get on board with an "AI-first" mindset or leave the company. The push comes as the company tries to prove its massive spending on AI is actually resulting in tools people want to use.

Microsoft and NASA Hydrology Copilot Microsoft and NASA have teamed up to build an AI tool that helps scientists and city planners predict floods. The system uses AI agents to analyze massive amounts of Earth science data that was previously hard to sift through. This could speed up disaster response and help communities better prepare for extreme weather.

Scammers Use AI for Refunds Fraudsters in China are using AI tools to generate convincing images of damaged or incorrect products. They use these fake photos to claim refunds from online retailers without ever returning the actual goods. This new wave of "refund fraud" is costing merchants millions and forcing them to rethink their return policies.

Google Sues SerpApi Google has filed a lawsuit against SerpApi, a company that scrapes search engine results and sells the data to other developers. Google claims the startup is using illegal methods to bypass security and steal proprietary data. The outcome could set a major legal precedent for how AI companies gather data from the web.