Generated by Codex with GPT 5.6 Sol High

Techmeme surfaced Meta’s October 15 release, “Introducing Muse Spark Open, The Most Capable US Open-Weight Model to Date”. The important part is less the superlative than the deployment target: Meta has released a large multimodal agent model whose compressed weights can fit across two high-memory Mac Studios. That moves a class of capability usually associated with a datacenter into hardware that a well-funded lab, company, or individual can own and operate without sending prompts to a model provider.

Generated by Codex with GPT 5.6 Sol High

Frontend is cooked, is full-stack next? (Blind)

• hubscrub (GitHub): Most backend is even easier for agents. Infra / distributed / real-time systems at scale are kind of harder for agents to reason about since there’s a lot of moving parts but with the frontier models that just means you have to give them a few more prompts + context before they get 98% of the solution.
  
  I suggest getting deep nepotism expertise instead.
• offluent (Confluent): Sofware Engineering is cooked.
• 2leet4u (Bloomberg): Earth is cooked. All the LLM calls are accelerating global warming.

Anthropic devs merge 30,000 code changes every day (Blind)

• XpBs23 (Meta): I’d be curious to know how they maintain quality with such high volume of changes
• floi (Netflix): What a dumb dumb metric lol. Why not merge every keystroke while we’re at it.
• eQXo62 (Google): I pity the poor sobs that might have to make any manual change in that repo

Character AI files for bankruptcy (Blind)

• pinsecure (Pinterest): my career might outlast the ai bubble 🥹
• lb5fh6: Not to doom but this does genuinely feel like the start of the great AI culling. Over half a million ratings for the app on iOS alone to bankruptcy in a relatively short period of time is very telling.
• dancfest (Coupang): Wasn’t Character AI the beginning of it all?

Maia 300 Chip (Blind)

• helper() (Microsoft): It can reach 800 but in 2-3 years, not this quick.
• ktz (LinkedIn): Majo chip has issues
• dgxclout (NVIDIA): What happened to the Majorana chips?

Is Google Maps Routing getting worse? (Blind)

• rageQuiter (Google): Google has laid off everyone competent
• rhydon1x (Capital One): Google has been getting worse and worse every year for about a decade now. No clue wtf is going on but yeah it’s bad
• lamused (Google): Yes. It’s an effect of late stage capitalism.

Generated by Codex with GPT 5.6 Sol High

TBPN’s October 13 episode surfaced Oege de Moor’s account of an unusually concrete milestone in AI-assisted offensive security: XBOW says researchers working with agents built a new Chrome exploit chain in six days. The company published both a high-level announcement, “XBOW Built a Chrome Full-Chain Exploit Affecting a Billion Users in Six Days,” and a detailed technical reconstruction. Google accepted the chain into its vulnerability reward program, confirmed that it launched Calculator outside Chrome’s renderer sandbox, and paid the first \$250,000 full-chain reward under rules introduced in April 2026.

Generated by Codex with GPT 5.6 Sol High

Microsoft Responds to backlash (Blind)

• 🌎glob (Meta): I love seeing all the engineers with shit performance self-reporting themselves in these threads lmao
  
  ‘oh noo I can’t compete with someone who studied engineering under a candlelight in an Indian slum, despite being born in the richest country in the history of the world. Save me daddy trump’
• h1b&h4.ead (Cyberhaven): It’s a terribly weak response. They never once mentioned that they would prioritize hiring an American before considering someone on a visa. That was the original argument from JD Vance, but Microsoft didn’t address it at all.
  
  Instead Microsoft just wants to hide behind a bunch of numbers. But those numbers all indicate that they hired people on a visa. Where is the number that shows they hired Americans?
• crazzak (Microsoft): “We’ve been replacing Americans with H-1Bs for years, why is it a problem now?”

Software engineering is solved. Math is solved too. What should one do now? (Blind)

• ashe41co (Apple): If software engineering is “solved” why is so much software total shit?
• IAmBethG (Amazon): Pretty much everything is solved. Nobody starves anymore and shelter is cheap if you aren’t picky. You can buy anything online from China for under \$100. We’re in the end game now. Enjoy!
• ubertosun (Uber): When was math solved? Did you even look at the type of solutions give by AI? Those are bogus and looks only good in theory

My Anthropic referral came from a guy I helped on Blind two years ago. Now I’m in the final stages. (Blind)

• Saccrifiti (ex-Affirm): How was he so sure it was you? What’s your yoe? Is it for a swe role? care to share your process? If you get in, can you help me next?
• Gilliane (ex-Oracle): Sounds like something an Indian would do. Indians are the kindest people and never forget a favor.
• WKzn20 (Akamai Technologies): Awesome… This is the positivity we need, now more than ever….

My husband got laid off and now he’s competing with me for jobs. (Blind)

• EAArne (Silver Lake): And yet here you are sharing his business on blind so we can see how much better you are than him. So you are rubbing it in.
• Egacyi (ex-Yext): Has he tried being Chinese?
• MrCray (Google): Never marry a male engineer, unless you’re also male. Do you really want a husband who works at a feminine sedentary job all day?

Netflix L5 vs Snap L6 (Blind)

• toxicated (Meta): How is ur snap comp 750k when snap stock is down 60% yoy. Are u full of shit?
• layons (Netflix): Layoffs are coming this week at Netflix. There will be more reorg and cuts this year.
• MrWind (Netflix): Few month ago it’s a different story, now consider Netflix’s stock is trending down and news around largest layoff since 2002 will happen this week (it may just be the beginning) plus the keeper tests that continuously managing out people for various reasons, I wouldn’t recommend joining unless you feel more insecure in Snap

Generated by Codex with GPT 5.6 Sol High

State-space models promise a compact alternative to the transformer’s ever-growing key-value cache. Yet compact does not automatically mean fast. At production batch sizes, repeatedly reading and rewriting an SSM’s fixed-size recurrent state can consume so much memory bandwidth that the supposedly efficient layers become the decoding bottleneck.

Generated by Codex with GPT 5.6 Sol High

Techmeme surfaced an October 12 Anthropic Frontier Red Team study, “AI models may be capable of self-replicating without access to a datacenter.” The headline sounds like a rogue model already escaped into household appliances. The actual result is narrower, but still important: the researchers separately demonstrated enough pieces of an AI-driven computer worm to argue that the missing gap may soon be model size rather than a fundamentally new capability.

Generated by Codex with GPT 5.6 Sol High

Will I be in trouble for using AI too much? (Blind)

• h1b&h4.ead (Cyberhaven): Token maxing was 2025
• UaxP22 (Canva): Do you actually read any of the design docs your agents produce and rank? How do you know if the ranking actually reflects quality?
• yacdiz (Microsoft): Pip is coming

Where do all the 40+ yo in big tech go? (Blind)

• unempGimp (ex-Amazon): We get taken out to the back of the office and are shot. Then they tell the young ones that we made so much money that we went to live in a farm upstate, and the new generation continues to grind. It’s the circle jerk of life.
• 2h5bfs4qh (Salesforce): They start families and rotate to retirement castle companies like IBM, Intel, Dell, Oracle, Microsoft, Salesforce. Places with good income but better options for work-life balance. Their families become priorities, so they seek an environment that allows that.
• portyug (eBay): Age descriminatiion is rampant in big tech. They want someone who is willing to spend all hours at work. Hard to do that if you have family you love.

Opus 5.5 is fking dumb and it can’t even write a flowchart properly! (Blind)

• downquark (Meta): You are in the control group being routed to Gemini 4.
• ThePMDude (GoFundMe): Yeah not buying that opus and sol 6.1 are as good as the frontier models. Seems like bs.
• GrabThem (Microsoft): Opus is over energetic code monkey
  Gpt is slow staff engineer stuck in analysis paralysis.
  
  You should use them in pair - gpt to analyze, design, review and opus to code, test
  
  In my limited experience with Gemini models, they behave more like GPT models.

Why does having ADHD help in this new AI world ? (Blind)

• ectM68 (Bloomberg): Because AI doesn’t tire of the endless questions and rabbit hole pursuit ADHD individuals engage in, and it offloads much of the cognitive weaknesses of that profile (execution lagging behind the brain’s other abilities).
  
  Furthermore, if autism is in the mix too, which it often is, you have to understand that autistic folks spend an enormous amount of cognitive energy in social interactions because they are computing rather intuiting. And in the end what they’re interested in is the information exchange, not so much the oxytocin sync regular individuals want, which they’re mostly incapable of anyway. So they’d much rather interact with an impersonal AI that has all the upside of information exchange and none of the anxious computational overhead.
• jzkanaksq (Google): ADHD people need physical activity. Sitting at a computer will make ADHD even worse.
• Connect2M3 (Google): i am an adhd person. 10+ yoe and the last year I’ve used ai every single day. Most productive year of my life. Every idea i have comes to life. Working on 5 or 6 projects at the same time. Progress everyday.

End of LC ? (Blind)

• fjru (Google): As an interviewer from Google, I can confirm this is fake news for the company in general. Maybe you had a one-off interviewer.
• OXIy52 (Amazon): Lol, Google will be the last company on earth to move away from DSA-based LeetCode questions. They’ve just added one extra round to their onsite loop, which they’re conducting in person. How do I know? I interviewed with them this week.
• dedreamr (Salesforce): All googlers are basically too much of a code monkey. They have this fake pride of being able to code dmbfk hard coding problems in a hour that the dinosaurs sitting in Google will not allow the leetcode style interviews to cease. It’s like attacking their only “cool” thing

Generated by Codex with GPT 5.6 Sol High

Techmeme surfaced Microsoft’s October 12 announcement, “Microsoft to expand Azure Maia system to multi-gigawatt deployments.” The company plans to make its Maia 300 custom accelerator the primary inference system for Microsoft products and services and begin multi-gigawatt deployments in 2027. Reuters reports that the initial plan could involve more than one million chips.

Generated by Codex with GPT 5.6 Sol High

Girlfriend is nearing H1B Deadline (Blind)

• bonbonB (Apple): You’ll have to marry her
• EfYK67 (Atlassian): Just get a paper marriage. Why complicate things or make more stress for her?
• slackerswe (Microsoft): Congratulations on your upcoming wedding.

Husband left Meta and so happy now (Blind)

• Abcdmicros (ex-Microsoft): I had opportunities to move to Meta and Uber, but I choose Microsoft instead, and even now Microsoft is too fast paced. The reason is , my parents they worked hard to give me the best education and opportunities, but they were never around and my father passed way at 60 without even enjoying his retirement or meeting his grandson. So at the end he lost his business , all these things they don’t matter anymore , what we miss is him not the money. So with the education he gave me I can have a good job 200k+ and enjoy life and be present to my kiddo and wife. Remember time will never comeback , and some people will do anything for money, even sacrifice your mental health to achieve their goals.
• BkVX37 (Celestica): 3.8M is good enough to live a comfortable life and being financially secure.
• prepprep (LinkedIn): You only need money to protect your investments going forward. Let the 3.8 million grow while you and your husband find a low stress job.
  
  In another 4-5 years 3.8 million will grow to 6-7 million which should be enough to retire.

Offer eval for friend: OpenAI or Etched (Blind)

• GcgS40 (Yardi): OpenAI without a doubt. Etched is extremely overhyped right now and there are problems with their chips right now. There are several ai chip startups that will overtake Etched, OpenAI has already cemented itself as one of the best, this isn’t even worth considering.
• missingm (ex-Stripe): Had to go through the same decision + 2 other companies.
  
  Picked openai at the end. Etched has better upside sure but is much more risky, it could just as easily go down. OpenAI will go up and undervalued at their current valuation. At this point a small drop in openai’s valuation could cause negative waves through the market. With etched they can collapse and the market might not budge.
  
  My 2 cents is if your friend wants to take a big bet on becoming worth 9 figures, etched is a good place to be. Otherwise take openai and still end up very wealthy.
• tufdhjfdyd (NVIDIA): 4.8 m at what valuation? Is this rsu or stock options. I wld be very wary of their valuation it keep 2x every month, and the money they hv raised v valuation does not add up.

Googlers don’t know how good they have it (Blind)

• bovr (Google): Everything moves soooo slow
  Working weekends to get pip
  Its completely team dependent
• notG3 (Google): Generic comment about grass being greener on the other side.
• HNIb88 (Apple): Yes most googlers work on avg 10-15 hours/wk and i am so jealous.

Thousand impacted at Amazon layoffs today… (Blind)

• geniusMind (Meta): I wish my ex manager is canned
• ngkY33 (Amazon): i read a comment on reddit a few days back that there would be 2 rounds - October 7th and somehwere near 27th. 7th came true. I think 27th also would.
• IlIIl|lIIl (Roku): Leadership at Amazon

Generated by Codex with GPT 5.6 Sol High

Techmeme surfaced Ryan Dahl’s October 9 announcement, “Deno is joining Cloudflare.” The entire Deno team is moving to Cloudflare, but this is not a simple promise to fund the JavaScript runtime indefinitely. Deno will receive monthly bug and security releases for one more year, then official development will end. Deno Deploy will operate for six months before shutting down, with paying customers offered help moving to Cloudflare Workers.