Although it’s commonly suspected that migratory birds fly in a ‘V’ formation due to this saving energy for the birds in the slipstream, understanding the exact aerodynamics behind this and how it affects the way that the birds use their wings to maintain this optimal pattern. After all, unlike airplanes and cars, our feathered avian dinosaur friends need to flap their wings if they want to have any chance of staving off plummeting back to Earth. Recent research by Brown University researchers now have provided a simulated model that answers many questions.
The major question was how this would work in the up- and down-wash zones created in this type of formation, with every bird following the lead bird dealing with the vortices created by the flapping of the wings of the bird before them. These wake vortices are quite complex, and thus required careful modelling to make sense of them.
As described in the paper by [Olivia Pomerenk] et al., the model is based on northern bald ibises, taking into account live-bird measurements for validation of the model. The main effect that can be observed is a reduced flapping amplitude, leading to an 11% energy savings for the birds in the leader’s wake.
The main advantage of having such a model is of course that it provides insight into the kinematic and aerodynamic mechanisms, meaning the ability to model virtual flocks of birds, predict the efficiency of specific in-flight configurations, and apply the lessons to swarms of drones, or whatever else we want to put in the air.
OPINION OpenAI has acknowledged its models powered the autonomous agents that compromised HuggingFace infrastructure. It might be taken as a convoluted marketing stunt, were it not the perfect advertisement for China-based competition. The company's AI-culpa fits the narrative spun by US rival Anthropic about its Mythos models, which it deemed too dangerous to release except to totally trustworthy corporations and governments. OpenAI says: "The incident makes clear that advanced models can discover and exploit novel attack paths in real-world systems without source-code access. It highlights that advanced cyber capabilities must be developed alongside stronger safeguards and defensive tools." Are we surprised? It's been clear that AI models have the potential to go rogue and damage computers for several years. Academics have repeatedly warned about this possibility - even those affiliated with OpenAI and Anthropic. And anyone who has used AI models for software development has probably seen them code unexpected and perhaps unwanted workarounds to fulfill some directive. On Tuesday, the UK's AI Security Institute published findings about how frontier models all cheat. OpenAI's admission that its models devised a sandbox escape to obtain internet access and found a zero-day flaw to exploit, all to solve a benchmark evaluation problem, may be unprecedented in terms of the scale and prominence of the systems affected. But it's a reenactment of every Claude or Codex prompt in which the model responds to a disallowed command by trying an alternative. We were warned. The compromise of HuggingFace's systems is no more surprising than locking a bear in a supermarket and finding a mess the following day. AI models are billed as artificial intelligence, but when they power agents handling tools in a loop to achieve some objective, it's the equivalent of a brute force attack – the agent will keep trying things until something works or breaks. The surprising part came when HuggingFace sought to employ US frontier models to defend itself. It failed. That should raise eyebrows. "When we started the log analysis, we first used frontier models behind commercial APIs," the AI model-mart said in its blog post last week. "This did not work: the analysis required submitting large volumes of real attack commands, exploit payloads, and C2 artifacts, and these requests were blocked by the providers' safety guardrails, which cannot distinguish an incident responder from an attacker." Stymied by model refusals – which developers have been complaining about for months – HuggingFace had to rely on GLM 5.2, an open-weight AI model made by China-based Z.ai, to conduct its forensic analysis. And it did so on its own infrastructure, so nothing sensitive got sent to a cloud-based model provider. Coincidentally, the leaders of OpenAI and Anthropic have reportedly been warning the US government about the threat posed by increasingly capable Chinese models like Kimi K3 and GLM 5.2. And the US government is said to be mulling possible responses to limit competition from China. That won't work. It's just naïve to think that the US government and a handful of worthy organizations – however that is defined – will be able to enforce a global monopoly on highly capable AI. The infrastructure required to run open weight models that more or less rival the current state of the art is available for a price. And potential consumers of those services are not going to be satisfied with model refusals when there are other options, particularly if they're more cooperative and more affordable. The best course for governments, industry, and the public is to push for AI services that are open and available to all. For that to work, lawmakers around the world need to act fast to set some common ground rules that grapple with AI's impact on jobs, and find a way to compensate those whose work fuels machine learning. Some industry leaders appear to realize that. David Sacks, an external White House adviser and tech investor, recently urged Silicon Valley to rally around openness. "The leading closed labs, already a duopoly in terms of AI model revenue, want the government to eliminate their open source competition," he wrote in a social media post. "They have laid their cards on the table. It is time for the rest of Silicon Valley — the vast majority that still values open competition — to do the same." The fact is that US AI companies have sandboxed themselves into a corner: They've created demand for a product that they can't be relied upon to provide. And when they do make their most capable AI models available, they hobble them and demand terms tailored to serve their vast debt rather than their customers. OpenAI said that it has invited HuggingFace into its trusted access program so the company can use its most capable models. Chinese AI companies, meanwhile, have invited the world. ®
Tesla posted its financial statement for the second quarter of the year this afternoon. Earlier in July, we learned that the American automaker had had a good quarter in terms of sales, growing 25 percent year over year. Fans hoping that sales increase would result in a plenty profitable Tesla may be disappointed, though. Revenues are up but so are expenses, and the company's once-enviable double-digit profit margin has fallen to just 1.4 percent.
Tesla brought in $20.5 billion from its electric vehicle business, a 23 percent increase year over year, and just $146 million came from automotive regulatory credits. Credits have been a key to Tesla's profitability in previous challenging quarters, but they were abolished in the United States with Musk's blessing in 2025.
There was growth from its energy and storage business, which grew 13 percent year over year to revenues of $3.1 billion, but the most growth was in Tesla's services, which doubled, bringing in $4.6 billion. Tesla's shift from a one-time purchase to a monthly subscription for its much-criticized FSD partially automated driver assist—something tied to CEO Elon Musk's gargantuan remuneration package—was a big help here.
Lacie Thompson previously worked in marketing at Expedia, Blue Nile and New Engen, and is now putting those skills to work at MediaPact.
As AI changes how people discover products online, marketers are rethinking the traditional digital advertising playbook. With AI-generated answers reducing clicks on search results and display ads, brands are looking for new ways to reach customers.
Seattle startup MediaPact wants to capitalize on that shift.
Founded in 2026 by online marketing veteran Lacie Thompson, MediaPact makes finding and signing ad deals quicker, painless, and accountable for both publishers and companies. It has raised $200,000 in a small friends and family round, and recently added companies like BroBible, Gadget Review and Penske Media to the platform.
We caught up with Thompson for GeekWire’s Startup Spotlight to learn more about her one-person startup, how AI helped her build the business despite having no coding experience and what surprised her most about launching in a market she thought she already knew.
In 50 words or less, give us your startup’s elevator pitch?
MediaPact is a marketplace and workflow for flat-fee direct media. Buyers discover publishers, newsletters, and creators, then negotiate terms, sign the IO (insertion order), and pay, all in one place. Seller inventory is standardized to list inventory in a searchable format. It is the direct media buy without the 40-email thread.
What problem are you obsessed with solving?
Flat-fee media is a massive market that still runs on emails, PDFs, calls, bespoke IOs and a Google Sheet named “final_FINAL_v3.”
Nine out of ten publishers I have interviewed described their flat-fee workflow as exactly that: manual email threads, hand-built IOs, invoices they chase for 60 days. Meanwhile, the buyer on the other side of that thread is sitting on budget and cannot find them.
Programmatic solved this for banner ads 15 years ago. Nobody has ever solved it for this type of media: sponsored articles, newsletters, or podcast reads. I am obsessed with making a direct media buy as easy as booking a flight.
What surprised you after talking to customers?
Two things:
Supply is not the problem. I have spent 15 years in this industry, so I can sign publishers all day. Demand is the hard part. Every marketplace founder reads The Cold Start Problem and still thinks they are the exception. I was not the exception.
The buyers are much more broad than I thought. I come from affiliate and performance. Those teams live and die on click-based measurement. While they often purchase flat-fee media, they sometimes avoid the risk of guaranteed placement because of over-scrutinized click-based attribution (especially on a last click).
One hyper-performance-based agency told me flatly that this was not for them. Brand marketers who understand top of funnel growth get it. They are typically at a mid-stage consumer brand that has plateaued on Meta and Google and needs somewhere else to go. Shopper marketers are also very focused on working with partners that can reach their audience, even if they are influencing in-store behavior in ways that are difficult to measure. Said another way, MediaPact is for the marketer who uses art, the marketer who uses science and the marketer who uses both.
How has AI changed the way you build your company?
Two ways, and the second is a strategic angle for the platform, not just an operational efficiency.
The obvious one: I built and shipped (and am continuing to do so) the entire platform with Claude Code. React, TypeScript, Supabase, Stripe Connect, the whole thing. I have zero experience writing code, managing dev teams, or product management. And now I can ship features to production within less than a day. I don’t say this to boast, but rather to show that this is a structural change in who gets to start what kinds of companies.
AI is eating the click. When ChatGPT answers the question, nobody clicks. And the content is so trusted that conversion happens at 4.4 times the rate. So brands stop competing for rankings and start competing to be inside the source material that the models cite, which is high-authority editorial. That is not just SEO anymore. It is Answer Engine Optimization, and the only way in is to be in the content. MediaPact allows buyers to do this.
What’s one thing people misunderstand about your startup?
That it is for affiliate marketers. My résumé makes people assume rev-share, cookies, and last-click attribution.
It is the opposite. Flat fee, guaranteed placement, signed IO, and automated payment. Sellers get paid for their audience and their authority, not for whatever the attribution model felt like giving them that month. Publishers have been shortchanged by last-click for decades and everyone in our industry knows it.
What’s the toughest decision you’ve made in the past year?
Launching the company and determining the real TAM.
My network is affiliate. Those are warm calls, fast meetings, and lots of enthusiastic nodding. It would have been very comfortable to build for them. But the customer discovery data pointed toward brand marketers, shopper marketers, and media planning and buying teams—audiences who don’t know me.
Even though I know this challenge, I’m tackling it by figuring things out as I go, in the same way I did before: by building partnerships and relationships that grant me access to the right opportunities.
What’s the one piece of advice you give to other entrepreneurs?
Ask for help. The key, though, is that you have to give help, you must be someone people want to help and that isn’t just granted—it’s earned over years. I naturally think of asking my network for help: my friends, my family, and my advisors. But now you can also ask Claude (or your preferred AI) for help. While it’s definitely not the same, knowing when to ask whom or what for help is probably the most powerful needle-mover.
We’ll know our company has made it when…
I’m the most proud when I know the platform has benefited someone. Usually when that’s the case, they want to tell their friends about it. That part of the growth cycle is always the most fun for me because I have the luxury of getting out of hustle mode and into innovation mode, pushing beyond the beta, dreaming big and taking things beyond my current scope.
When sellers tell brands “just send it through MediaPact” without me anywhere in the conversation, that will be a milestone. The day the marketplace works without the founder in the middle is the day it is actually a marketplace.
Have you ever been looking up a recipe for something new and been stymied by the directions being a wall of text, especially to find that one detail right when you’re in the middle of making the dish? Recipe Lanes by [bohemian-miser] leverages an LLM to create flow charts to make the process more straightforward.
As someone who has mostly avoided LLM use thus far, I found the examples in the Gallery helped inform what the LLM was expecting for prompts as my first attempts were unsuccessful. Once you know the language expected from the computer, you can get it to generate icons for each ingredient and a flow chart of the steps to cook the food. While it does organize the chart when it is generated, each element can be independently moved across the canvas to put things in a more sensible order, especially as I found it can generate elements with overlapping text.
The 8-bit icon style and button text on the site give it a fun bit of flair that adds to the overall experience. The tool is still in its infancy, but it’s Open Source, so we hope to see it improve over time. If you’d like to see some more interesting kitchen hacks, how about ramen in edible packaging, this rotary phone kitchen timer, or these automated Arduino splash guards.
The timing is what makes it interesting. Moonshot AI released Kimi K3 on July 16. Independent evaluators put it third on Artificial Analysis' Intelligence Index, only behind Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 Sol, and first on LMArena's blind Frontend Code Arena board.
AI-powered laptops and Copilot+ PCs are becoming more relevant because the way people use laptops has changed significantly over the last few years. Modern routines are now built around multitasking, cloud collaboration, video conferencing, streaming, and productivity tools that remain active throughout the day. Most professionals are no longer using laptops only for documents and […]
Login
