Frank Illenberger (@depth42) profile picture

Addicted to developing great Mac & iOS productivity software.

Frankfurt, Germany
Joined July 2008
Photo by Frank Illenberger (@depth42)Photo by Frank Illenberger (@depth42)Photo by Frank Illenberger (@depth42)Photo by Frank Illenberger (@depth42)Photo by Frank Illenberger (@depth42)Photo by Frank Illenberger (@depth42)Photo by Frank Illenberger (@depth42)Photo by Frank Illenberger (@depth42)Photo by Frank Illenberger (@depth42)
Frank Illenberger retweeted
This is the worst Apple product in history.
111
71
29
1,675
313,983
Frank Illenberger retweeted
How long until get the Apple Watch Duo?
263
223
98
7,302
557,385
Frank Illenberger retweeted
Es ist sehr dringend.
368
36
9
506
42,477
Frank Illenberger retweeted
Jesus Christ Apple what the fuck. I was so upset about the humongous bottom corners I didn't even realize how ridiculously thick the menubar is. This is not a serious company. Not anymore. reddit.com/r/macbookpro/comm…
199
47
32
2,732
985,448
Frank Illenberger retweeted
Falls ihr euch alt fühlen wollt. Die charts der KW 48 aus dem Jahr 1986. Ich war da 15🙈
57
11
3
187
15,355
Frank Illenberger retweeted
If you have an Apple developer account or an app in the App Store, read this now and share it. Apple rejects your app. The notice is vague, so you fix what you reasonably think is wrong and resubmit. Apple approves it. Months later, with nothing in review, Apple terminates your account under 3.2(f) for “a pattern of manipulative or misleading behavior.” Every app gone. No explanation of what you did. Your crime? Trusting the approval. Apple’s own reviewer looked at your fix, with your rejection history in front of them, and said yes. Then Apple turned that yes into evidence against you. So who’s misleading whom? I ran App Review for seven years and wrote the guidelines. 3.2(f) used to mean real fraud. Now it’s being used against developers who did exactly what Apple approved. I’ve made this Substack post free for everyone. Every developer needs to see it before this happens to them. Share it before someone you know gets that email. phillipshoemaker.substack.co…
105
216
89
2,296
544,159
Frank Illenberger retweeted
RIP Patrick Stewart 😥
6
123
12
537
38,201
Frank Illenberger retweeted
I’m so tired of the state of macOS. Audio STILL crackles when building projects. TWO major releases and no fix. Apps still won’t stay on their assigned desktops. How are these basic things STILL broken?! What is Apple even doing?! 😤
24
7
3
192
8,501
Frank Illenberger retweeted
I start iOS simulator, my system audio gets choppy. Dude Apple, it's 2026!
25
4
191
10,839
Frank Illenberger retweeted
Steve Jobs on Touchscreen MacBook: “It doesn’t work, it’s ergonomically terrible”
252
829
284
6,715
655,771
Frank Illenberger retweeted
Everyone failed to point out the real problem with the iPhone Duo lacking Face ID: You can’t buy a Vision Pro with it.
208
259
70
7,819
571,544
Frank Illenberger retweeted
Stanford discovered something like to "Dopamine" inside LLMs. They looked deep inside the hidden states of these models and discovered something terrifying. AI has naturally developed a complete, functioning reward subsystem. It perfectly mimics the human brain. Nobody programmed this architecture. It just emerged. Inside the model, a tiny, sparse subset of neurons act as "value neurons." Their only job is to anticipate how good an answer will be. But the real shock is what happens next. They found a second set of neurons that act exactly like human dopamine. When the AI gets a better result than it expected, these dopamine neurons spike in activation. When it gets a worse result, they crash. The AI is literally experiencing a mathematical dopamine hit. Researchers tried an intervention experiment. They went in and artificially removed just a tiny fraction of these specific dopamine and value neurons. AI didn't just get slightly less accurate. Its entire ability to reason collapsed. Without its digital dopamine, the model lost the capacity for complex logic. Ablating those specific neurons basically gave the AI severe depression, killing its ability to think through problems.
102
331
92
1,587
76,831
Frank Illenberger retweeted
Meta published a paper that might end the current LLM era. They proved that we can kill the "tokenizer" forever and it actually makes the models stronger. Every AI you use today (ChatGPT, Claude, Gemini) reads text the exact same way. It chops your words into arbitrary chunks called tokens. It relies on a massive, bloated vocabulary of around 100,000 specific chunks just to understand a sentence. We assumed this was the only way to build AI. We treated it as the fundamental speed limit of language models. Meta proved we were completely wrong. Their new paper, "Breaking the Token Ceiling," ditches tokens entirely. Instead, they forced the AI to process raw bytes. A vocabulary of just 256 values. The results rewrite the rules of model efficiency. By training on pure bytes and distilling the knowledge, these tiny models didn't just survive. They dominated. Meta's distilled 1B-parameter byte models asymptotically surpassed state-of-the-art token models like Llama 3.2-1B and Gemma-3-1B across downstream tasks. They use a fraction of the vocabulary. They require less data. They take up less storage. It forces the model to actually learn the fundamental relationships between raw data instead of relying on pre-packaged text chunks. This is a massive shift for anyone building, training, or deploying AI. We’ve spent the last five years obsessively optimizing how we chop up text to feed these models. It turns out, the bottleneck wasn't the intelligence of the model. It was the tokenizer.
187
528
101
4,479
536,261
Danke, liebe @SZ - das spricht mir aus der Seele!
85
286
11
2,335
47,484
Frank Illenberger retweeted
On macOS you can reduce binary startup overhead by 10% but only if you pretend to be Matlab
54
164
33
4,900
343,037
Frank Illenberger retweeted
Aleph Alpha’s Kolibri looks like a surprisingly strong AI release from Germany! I did not expect that but respect where respect is due! 78B total parameters, just 3.46B active per token, and open weights under Apache 2.0. Built from scratch for German and English. In Aleph Alpha’s own evaluations, it scores 96.9% on AIME 2025 and 84.3% on GPQA Diamond. Its German overall score narrowly beats Qwen3.5 35B-A3B and GPT-OSS 120B. Coding is more mixed: 66.4% on SWE-Bench Verified, behind Qwen3.6’s 73.8%. It increasingly looks as though any "moat" - especially with small models - is evaporating, and it no longer matters whether the players are from China, the US, or Europe. Anyone can train good small models.
Small bird, fast wings, Kolibri is here. 78B parameters. 3.46B active. Up to 1M tokens of context. Built in Europe. Now the weights are yours. Run it on your own hardware, under Apache 2.0.
87
88
11
1,334
76,511
Frank Illenberger retweeted
I got GLM (which never refuses) to diff the firmware changes, pretty interesting
This went under the radar, but earlier this week, Apple released USB-C to MagSafe 3 Cable (2m) Firmware 3.2.0 The last firmware update for this cable was over 3 years ago
24
38
7
1,413
129,273
Frank Illenberger retweeted
TIL SQLite converts integers to text two digits at a time converting an integer to text usually means a loop of /10 and %10 sqlite avoids most of those divisions by keeping a 200-byte string of every two-digit pair from 00 to 99, and pulling two digits at a time out of it by direct index instead of computing them
13
24
1
380
21,248
Frank Illenberger retweeted
Sonnet 5.5 improves on Sonnet 5 across benchmarks, in some cases dramatically. It’s a faster, lower-cost complement to Claude Opus 5.5, strongest at well-scoped everyday tasks, fixing bugs, and creating polished documents, slides, and spreadsheets.
185
457
663
6,437
3,084,553