🍭 Your Next Job Interview Might Be With AI 🤖

They Weren’t Kidding About the Pace of AI.

Good morning. People working behind the scenes had been watching it build for years before ChatGPT arrived in 2022. I remember a talk Darío Gil gave at MIT in 2017. His message was clear: IBM Research was going all in on AI.

Big bet vibes.

It sounded important sitting in that room. Then we walked out, life kept going, and everyone returned to their own research, deadlines and promising technologies. I wonder how many of us understood just how much that bet would eventually change our work.

For plenty of people outside the field, AI still felt distant. Something happening in somebody else’s lab.

Then ChatGPT arrived.

Hi, I’m Samuel, and I’m bringing you today’s issue. I got my PhD at MIT a few years back. I went to middle school with Eder, our founder. Sort of amazing that we’re here together today, but here we are. More on that later. I’m joining the Snack Prompt & Daily Bite team to help make sense of what this technology means for the work and lives we care about.

Right, about AI.

You know how the next part went. Cool poems. Cheesy songs. A few simple prompts, and suddenly everyone had something to show their friends.

Now we’re handing AI tools, files and actual responsibilities. Bots and agents everywhere, a Wi-Fi connection away. Doing work for individuals, small teams and large enterprises. The distance between “look what it wrote” and “look what it did” keeps shrinking.

Last weekend, two friends, a robotics PhD and a tech leader in photonics, told me they’re concerned about their careers. These are people who understand the technology. That stayed with me.

Maybe AI will give us more freedom. I hope it does. There’s plenty of talk about a future in which we won’t need to work. But I keep thinking about the work we’d still want to do.

Work can be a good thing. The people. The feeling of being useful. The satisfaction of serving someone, solving a stubborn problem, or finally getting that cool thing you built to work.

Then I read about an AI system conducting job interviews. My friends’ questions suddenly felt closer. How will we show what we can do when AI is helping evaluate us?

Now we are here. One of today’s stories: HackerRank’s AI interviewer is changing how candidates demonstrate their skills. Reflection’s new model, Beam, promises another powerful tool for putting those skills to work, with downloadable weights still to come.

Back in that room at MIT, going “all in” sounded like a research strategy. Today, the changes are reaching the work we do and the interviews we take to get it.

I’m hopeful. Learning a lot every day, but hopeful. I’m also listening. How are you feeling about the pace of AI? 

Enough reflection ❤️ Let’s dive in.

Today’s Menu 🥠

• The AI interviewer: HackerRank wants to see how you think

• Meet Beam: a new open source model, with downloads still to come

• Fast Snacks: Google’s power deal, DeepSeek’s funding and AI watermarks

• Everything else: three useful tools, one startup 👀, which image is AI?

Tuesday, October 6 · Read Time: 6 minutes

The Big Bite · HackerRank
🤖 The Interviewer Has Entered the Chat.

The Bite: HackerRank made its AI interviewer, Chakra, generally available on Monday. It asks questions, follows up and helps employers assess candidates. Yes, even the interview has an AI colleague now.

What’s different? Chakra can watch candidates tackle practical problems, including coding with AI, and probe their decisions. The pitch is to see how someone thinks, checks an answer and uses the tools available to them. A working result is only part of the story.

The detail to catch: HackerRank says its scores are tied to evidence in the interview transcript. Its documentation leaves review and hiring decisions with the hiring team. Those are design claims, not independent proof that every assessment will be fair or accurate.

Why it Bites: My friends are wondering how their expertise will count as work changes. This puts that question right at the entrance. Using AI well may become something you need to demonstrate before you get the job.

Our take: practice explaining your choices. Show what you asked the AI, what you checked and what you changed. If you’re hiring, read the evidence behind the score. A person’s opportunity deserves more than a number. And a personal take.

Second Bite · Reflection
🔑 Meet Beam. The Download? Not Yet.

The Bite: Reflection AI unveiled Beam on Monday, its first open model for coding, reasoning and agents that use software tools. It plans to release the weights so businesses can run and adapt the model themselves.

What’s different? Reflection claims competitive results against leading open models (a lot of them Chinese models) with less computation spent on reasoning. Independent testing will matter. With 501 billion parameters and 23 billion active at a time, this is a substantial model, not a casual laptop download.

The detail to catch: early access signups are open. The weights, technical report and model card are promised later this month, after final safety testing.

Why it Bites: Think back to that IBM lab. The research still takes serious resources. Releasing model weights can give other teams a starting point they can adapt and a version they can keep. That choice matters as AI becomes part of how we work.

Our take: when Beam is available, try it on a task your team knows well. Count the corrections, cost and time saved. Your own work is the benchmark that matters.

Fast Snacks 🥡

⚡ Google Orders a Lot More Electricity
Google and Constellation announced a 3.59 gigawatt power deal Tuesday, including 890 megawatts of additional nuclear capacity from upgrades to existing plants. The entire deal is not nuclear. AI’s next leap needs a place to plug in.

💰 DeepSeek’s Reported $12 Billion War Chest
Reports Tuesday put DeepSeek’s upcoming funding round at roughly $12 billion, with a close expected this month. The money is not confirmed as secured. If completed, it would give a major Chinese AI competitor substantial resources for the next stage of the race.

🔎 AI Text Gets a Quiet Signature
OpenAI announced text watermarking for eligible EU ChatGPT and Codex outputs over the coming weeks, plus an optional API setting. Editing can weaken the signal. A watermark can help trace an output’s origin; it cannot tell you whether the answer is true.

ToolBox™
🧰 Three Tools. One Startup to Watch.

🔗 Together Link: Connect coding tools such as Claude Code and Codex to open models hosted by Together AI. Launched Monday. Useful for comparing model choices without changing your working setup. API usage is billed through Together; test the cost and quality on your own tasks.

🧭 Liquid AI d1: Monday’s update adds vision to a model that returns probabilities for choices. Try sorting support tickets or flagging defects in product photos. Available through Liquid’s API and playground. Its striking cost comparisons come from six company tests, so check your own workflow.

🎨 Higgsfield Ads Studio: Build a brand kit from your website or enter one manually, then generate static ads for a product. Useful for exploring several creative directions before spending on distribution. Introduced October 1. Review the copy and credit cost before running a batch.

🏡 Startup watch: Rhem Labs. This YC Fall 2026 team is building connected hardware and a data platform for care at home. Early days. We’re watching how it handles a deeply human need: helping families care for aging parents. Product availability and outcomes remain to be verified.

Test Your Skill
Snack Quiz: Spot the AI Image

One of these is a real photo. One is AI. Which one is AI? Vote below.

Let’s see if your AI radar still works. Have the machines learned your blind spots? 👀