AI & LLM news 🤖 | Tech updates ⚡ | Apple • Google • OpenAI • Claude | Phones, Macs & emerging technology
Joined November 2021
- Tweets3.9K
- Following53
- Followers181
- Likes372
Amazon just told its engineers that token usage is a bad target.
Terk Terkowitz, a VP and Distinguished Engineer at Amazon, published a guideline for software engineers this week:
- Measure long-term customer outcomes
- Input metrics like token usage, AI adoption, lines of code and code reviews are "deeply flawed" as targets
- "Tokenmaxxing is at odds with frugality"
- If AI tooling doesn't help your work, document why. It's "a finding, not a failing"
- Get closer to customers
Fair point from the memo itself. For now those input metrics still correlate well with output, and teams that don't adopt AI are falling behind. So it's no excuse to stop using the tools.
Gergely reads it as bad news for engineers who don't want to be customer-facing. I partly agree. The memo rewards people who can explain why their change mattered.
In August, Codex burned 2.5B tokens on my account in 30 days, and no user ever noticed that number. In my opinion, tokens are a cost line. A huge agent diff also costs reviewer time. What I'd track is boring. What shipped, what broke, and how fast we rolled it back.
Treat tokens as a target and you just built a leaderboard for invoices.
ALT Grey tabby cat mascot in a teal hoodie thinking with a paw on his chin, holding one small parcel next to a towering stack of tokens.
Amazon's leadership is very good at writing, but you still need to read between the lines.
What I read between these lines:
Engineers who are not "product engineers" (aka eng + PM interest) and don't like to be customer-facing will not have a future inside of Amazon
Full: aboutamazon.com/news/company…
2
4
526
Astro Burn وصلت iOS بعد إطلاقها على Android
تلعب بقط طيار ضد الزعماء، مع تحكم باللمس معاد تصميمه
سعرها £9.99 على App Store
apps.apple.com/il/app/astro-…
pocketgamer.com/astro-burn/i…
155
Claude Code Ctrl+W :
- Ctrl+W deletes back to the previous whitespace, so one press removes a whole path like src/utils/foo.ts or a --flag=value
- Ctrl+U deletes to the start of the line and Ctrl+K to the end; both keep what they deleted
- Ctrl+Y pastes the text you last deleted with any of those
- Ctrl+_ undoes your last input edit and puts the cursor back
2
112
اختبار الـ router الحقيقي.. يقلل التأخير ويحافظ على جودة الإجابة
أرقام السرعة الإجمالية لحالها ما تحدد أفضل مزود لاحتياجك
We just shipped our biggest improvement to our provider router since the founding of OpenRouter.
Before, we prioritized mostly based on price. The new router now rewards the best quality inference for the actual cost of a request, so requests stay cheap but are also fast and reliable.
Providers that deliver more value automatically win more traffic, so every provider has a reason to keep improving. And every time a provider gets better, so does your inference on OpenRouter.
145
تدريب النموذج يحل نص المشكلة.. والـ harness يحدد وش يتعلم منه
يعني الأفعال والتغذية الراجعة تعتمد على الـ harness
how do you post-train an agent?
we laid it out as a 5-step curriculum, from SFT to RL through a real harness, with livestreams, code and docs for each step
huggingface.co/blog/huggingf…
82
Qwen 3.8 27b unleashed just got listed:
pirateface.co/outsourc-e/Qwe…
Run Q3. Its insane.
Magnets now on Pirate Face:
@Blackfrost_ai · Qwen3.8-27B-ABLITERATED-GGUF
@na0x2c6 · Sage-A1.5-GGUF
@outsource_ · Qwen3.8-27B-Unleashed-GGUF
@LeRobotHF · pi05_base
@liquidai · LFM3-1.7B +19
pirateface.co/activity?tab=s…
5
11
859
NEW VIDEO - Apple decided to drop a touchscreen Macbook Pro today. Let's talk about it: youtu.be/2reQut8WNGc
⚡ @OpenRouter tip #6
Add the header X-OpenRouter-Cache: true and OpenRouter saves your entire response. Send the exact same request again and you get it back in about a second, for $0.
This is different from prompt caching. Prompt caching only discounts the repeated part of your input, and the model still runs. This skips the model entirely.
So if you're iterating during development you can turn this on to call the model without needing to pay for a new generation every time.
The cache lasts 5 minutes by default, up to 24 hours with X-OpenRouter-Cache-TTL.
ALT TypeScript fetch call to the OpenRouter chat completions API for anthropic/claude-sonnet-5.5, labeling a support ticket. The highlighted line is the header "X-OpenRouter-Cache": "true", which turns on response caching.
Apple has raised the MacBook Neo’s education price from $599 to $649, leaving students with just a $50 discount 😬
The MacBook Neo originally launched for just $499 with education pricing back in March, meaning its price has increased by $150 in just seven months
Still worth it?
اختيار الصور تلقائيًا يوفر وقتك..
لكن اختيار الذكريات يظل شخصي، وI'd أفضّل تعديل الاختيارات بسهولة.
The Alexa Photo Frame can use the smart assistant to choose your best photos based on lighting, smiles, and the season. theverge.com/tech/1019626/am…
52
Firefox can now fill out boring forms and monitor websites for you
Mozilla is giving Firefox’s Smart Window two new tricks. If you haven’t tried it yet, Smart Window is an optional AI mode in Firefox with a built-in chat that helps you get things done while you browse. With this update, it can fill out online forms using details from your tabs and keep an eye on websites for you.
digitaltrends.com/computing/…
1
3
2,156
Astropad Workbench 1.4 يضيف طرفية Mac عن بُعد لتطبيق iOS
يمديك تتنقل بين سطح المكتب والطرفية من iPhone أو iPad وتكمل جلستك، مع نقل ملفات Mac-to-Mac ودعم iPhone Duo.
المجاني 20 دقيقة يومياً، والاشتراك $14.99 شهرياً أو $79.99 سنوياً.
apps.apple.com/us/app/id6758…
50
Flipcorder — your iPhone as a 1991 camcorder.
Shoot 20-second flips that stack onto one dated tape a day, VHS or DV style. No editing, no filters, no ads. $4.99 once.
Built for iPhone Duo — folding the phone IS the power switch. The anti-Instagram home video with the chatter left in.
apps.apple.com/us/app/flipco…
49
Casting used to be locked the moment you wrapped. Runway just made the cast editable in post, which quietly kills the most expensive line in video production: the reshoot.
47
Runway Ads — describe it, run it as a video ad
Creative testing used to be a slog: you ship a batch of variants, then sit around waiting for data to prove a winner. Runway Ads flips the script. Feed it your creative direction, and the platform spits out on‑brand video ads for every channel, instantly learning which cuts through and instantly iterating to beat the best. Teams like Lovable are already on board, proving the loop works in real time. The cost of finding the winning video ad? Practically zero.
If you can describe it, you can run it as a video ad.
Runway Ads turns your creative direction into on-brand ads for every platform, then learns what wins and builds the ad to beat it.
Teams like @Lovable already use Runway Ads to scale ad creative and cut production time.
41
App Studio is Runway's version of what Cursor and Grok Build do for code — turning a plain description into a working tool at zero typing cost. The difference is the output is a creative app, not a CRUD dashboard. That's the right specialization for their audience.
Now you can build creative apps in Runway.
App Studio turns a plain-language description into a working tool, powered by Runway’s models. Watch below.
44
Apple bundling Pixelmator Pro into Creator Studio alongside Final Cut and Logic is the quietest acquisition payoff in years. One subscription, pro-grade image editing that can finally compete with Adobe — and it ships with M6 Macs.
Final Cut Pro, Logic Pro, and Pixelmator Pro get major AI upgrades in Apple Creator Studio
From AI-powered image editing to professional video effects, Apple's latest update gives creators more ways to bring their ideas to life.
digitaltrends.com/computing/…
40
My dream AI agent tool:
- Opus 5.5 Level Model (never delegates to worse)
- Spins up cloud and local code sessions
- Dots level real time voice / tool use
- Codex level Computer use on local sessions
- Claude Artifacts quality decks, docs and design
- Mobile app is UI of duet
- Inline mini apps (What Grokbot is doing with email and slack, but do this with alot more tools)
- Webhooks (Like Grokbot has)
- Built in social media access across all platforms
- Plugins (of course)
In summary:
I want an always on, frontier model powered employee that can use its own computer, and can use my computer 24/7 that self organizes all of its tasks into channels so that its easy to keep track of all the things we've done together. And I want to be able to control this agent on mobile, web, desktop, and most importantly real time voice.
Ollama can now generate speech locally on Apple Silicon Macs.
Its first supported speech model family, Gradium's Phonon, streams audio as it is generated and supports five languages. Ollama says its MLX implementation generated speech 35% faster than Gradium's implementation in tests on an M5 Max MacBook Pro. It requires Ollama 0.40.11 or later; Windows and Linus support is coming.
9to5mac.com/2026/10/26/ollam…
15