Image: made using DALL·E (Custom edited by me)
RSN021
Published 18.08.2025.09.00.MON
Mesh Messages, Open Models & Browser Bids
Hey, it’s Ritik! A lot is moving in tech: Jack Dorsey launched a Bluetooth mesh messenger that runs without internet, OpenAI released its first open-weight reasoning models, and Perplexity shocked everyone with a $34B Chrome bid. Project ATOM carries multiple meanings, from AI to nuclear policy. All this while Musk and Apple are clashing once again over App Store favoritism. Networks, models, browsers, and platforms are all in flux this week, and I’ll cut through the noise with what really matters.
1. Bitchat Mesh: Jack Dorsey’s offline messenger
Decentralized chat without the internet.
What’s Happening: Bitchat Mesh is Jack Dorsey’s new short-range messaging app that talks over Bluetooth mesh, not the internet. Available on iOS (App Store) and Android (GitHub build), it creates ad-hoc “nodes” from nearby phones to relay end-to-end encrypted text messages. No login or account; you pick a changeable display name. You can post to a public channel (anyone in range) or do one-to-one private chats. Range is 10m/33ft per hop, but messages can traverse farther if enough devices bridge the gap—that’s useful in stadiums, festivals, planes, subways, or dead-zones where cellular data chokes. There’s a panic mode that wipes private chat history. Early reviewers praise the simplicity but flag missing features (no images/audio yet, no Wi-Fi Direct but Dorsey says it’s coming). Security posture looks promising (no PII, decentralized, E2EE), but no external audit yet, so avoid sensitive data for now. Competing mesh messengers (Bridgefy, Briar, Air Chat) have longer histories; adoption density will decide utility because more nodes = longer reach. Copycats are already in Play Store; I would stick to the official builds.
Why it matters to you: Bitchat can keep groups coordinated when service dies, think concerts, conventions, emergency meetups, or travel. It’s a low-friction tool to add to your “offline kit,” especially if you work events or live in spotty-coverage areas.
Why it matters to the world: Resilient, peer-to-peer comms reduces single points of failure and censorship risk. If mesh tools hit mainstream, communities gain a fallback channel during outages, disasters, and network throttling.
2. OpenAI’s ‘open’ reasoning models land: gpt-oss-120b & 20b
Open-source reasoning takes the spotlight.
What’s Happening: OpenAI released two open-weight reasoning models, the gpt-oss-120b and gpt-oss-20b under Apache 2.0 on Hugging Face. Both use mixture-of-experts (activating ~5.1B params/token on 120b) and heavy reinforcement learning post-training to improve tool use and step-by-step reasoning. They’re text-only but can call tools (search, Python) and forward tough queries to OpenAI’s closed models via a bridge. OpenAI claims these models outperform leading open rivals (e.g., DeepSeek, Qwen) on several reasoning/coding benchmarks, while trailing its own o-series. Caveats: higher hallucination rates than o1/o4-mini, and “open” doesn’t include training data disclosure. Hardware claims note 120b can run on a single Nvidia GPU and 20b on laptops with 16GB RAM, expect tight optimizations and trade-offs. Strategically, this is OpenAI’s first major open release since GPT-2, aligning with U.S. policy pressure for American-led open stacks and developer demand for locally runnable models.
Why it matters to you: You get capable, license-friendly models to self-host for private code, docs, or research and no per-token bills, easier compliance, and the option to bump hard questions to a cloud model when needed.
Why it matters to the world: A strong U.S. open stack counters concentration of AI power, accelerates academic/SMB innovation, and pressures proprietary vendors on price and openness, shaping the AI commons.
3. Project ATOM: many names, one theme—foundational power
America’s $100M open-source AI comeback.
What’s Happening: “Project ATOM” currently headlines as American Truly Open Models—a $100M push (10,000+ GPUs, university + industry backing) to restore U.S. leadership in open-source AI as Chinese labs surge. The bet: openness spurs safer, faster innovation and sets global norms. But “Project Atom” also refers to CSIS’s nuclear posture studies (since 2015) exploring future U.S. deterrence options out to 2050 force structure, escalation control, and multipolar risk. Internationally, Kazakhstan’s ATOM Project campaigns to abolish nuclear testing, drawing moral authority from Semipalatinsk’s history. There’s a U.S. nonprofit “Project Atom” focused on education equity for refugee and migrant youth. And in pop culture, DC Comics’ Captain Atom echoes scientific transformation. Across contexts, “ATOM” signifies the smallest unit with outsized consequence and policy blueprints, open models, disarmament advocacy, social mobility, or mythic science. The 2025 AI ATOM aims to make open models the foundational building blocks of digital progress, while nuclear ATOM reminds us that managing existential tech is as vital as creating it.
Why it matters to you: Expect higher-quality, open models, datasets, and tooling you can actually use without vendor lock-in. For builders, that’s lower cost, deeper control, and better auditability.
Why it matters to the world: Open AI leadership shapes standards on transparency and safety, while nuclear ATOM work shapes deterrence and disarmament. Both are governance plays over civilization-scale technologies.
4. Perplexity lobs a $34.5B bid for Google’s Chrome
Browsers are the new battleground.
What’s Happening: Perplexity AI made an unsolicited $34.5B all-cash offer for Chrome, vastly exceeding its own $14B valuation and saying unnamed funds would finance the deal. Alphabet hasn’t put Chrome up for sale and plans to appeal last year’s antitrust ruling; the DOJ has floated Chrome divestiture as a remedy. Other suitors reportedly circled (OpenAI, Yahoo, Apollo). Why Chrome? Browsers are the gateway to search, data, and “agentic” workflows; control there could rocket Perplexity’s AI answer engine. The offer is also a pressure tactic amid regulatory uncertainty: if courts force structural remedies, Perplexity wants a seat at the table. Practical hurdles abound: funding specifics, integration risk, Google resistance, global approvals, and whether a browser carve-out is even feasible without shattering Chrome’s ecosystem.
Why it matters to you: If any browser changes hands, your defaults, privacy settings, extensions, and search experiences could shift, and we can expect bundles of AI features, new ad models, and renegotiated data terms.
Why it matters to the world: A forced Chrome divestiture would redraw search distribution, ad markets, and AI discovery, arguably the biggest browser shakeup since Chrome’s birth, with global competition stakes.
5. Musk threatens to sue Apple over App Store “favoritism” toward ChatGPT
Chatbot wars spill into the App Store.
What’s Happening: Elon Musk says Apple’s App Store unfairly boosts ChatGPT in rankings and curation while holding back Grok (xAI). He called it an “unequivocal antitrust violation” and vowed legal action. Apple replied the App Store is fair and curated via charts, algorithms, and expert lists; Grok currently sits around Top5 free apps in the U.S. Context: Apple integrates OpenAI models into Apple Intelligence; Musk co-founded and later sued OpenAI and has threatened to ban Apple devices at his firms over that tie-up. Separately, critics allege Musk tweaks X to amplify his own posts and Sam Altman resurfaced those claims in response. Legally, proving discriminatory ranking is hard without discovery; Apple has strong defenses around editorial judgment and quality bars, but antitrust scrutiny of app stores is elevated worldwide.
Why it matters to you: App discovery rules who wins your screen time. If lawsuits or regulators force changes, you could see more transparent rankings, new default choices, or different pay-to-promote limits.
Why it matters to the world: Gatekeeper platforms decide which AI assistants scale. Policy outcomes here will shape competitive dynamics, privacy norms, and how quickly new AI entrants can reach the masses.
Creative Conclusion
From mesh texting without internet to “open” models you can actually download, this week felt like a tug-of-war over who controls the gateway, networks, models, browsers, or app stores. My take: resilience (mesh), openness (gpt-oss/ATOM), and real competition (Chrome bid, App Store fights) are healthy pressures. They won’t all land, but they’re steering us toward tech that’s more portable, auditable, and user-choice-friendly. Keep your tools diversified, your data local when you can, and your defaults intentional. See you next week.
—Ritik Sharma Founder, RSN (Ritik Sharma Newsletter)