Heelalle

Technology

technology

I have been working in cyber security for the last 15 years. Too much of thinking during 9to5. Can I work in bubble tea shop? I hope they don't pay well 

#sydneycbd #sydney

Yay just finished my  InfoSec Job - Sydney

Ek werk al 15 jaar in kuberveiligheid. Te veel om te dink tydens 9 tot 5. Kan ek in 'n bubble tea-winkel werk? Ek hoop hulle betaal nie goed nie 

#sydneycbd #sydney

Yay, ek het net my InfoSec werk in Sydney klaar gemaak

OMGMEDIA

The future is now. Today I had the opportunity and pleasure to talk to a humanoid robot. 😅

The future is now. 🤔👁👁

Die toekoms is nou. Vandag het ek die geleentheid en plesier gehad om met 'n humanoïde robot te praat. 😅

Die toekoms is nou. 🤔👁👁

Weekend Plans!

Hierdie naweek sal met QML deurgebring word

Naweekplanne!

A beginner's model for rainfall prediction.

First Draft Done!

'n Beginner se model vir reënvalvoorspelling.

Eerste konsep klaar!

set up

haai daar, deel asseblief jou werkopstelling

stel op

Why do people like the small talk so much

Hoekom hou mense so baie van klein praat

- I'm 44 in less than 2 weeks (July 3rd)
- I'm going into an android engineering career 
- I'll be moving in 2027 ✌️

Updates since I've been gone for 8 months

- Ek is 44 oor minder as 2 weke (3 Julie)
- Ek gaan in 'n android ingenieurskap loopbaan in 
- Ek sal in 2027 verhuis ✌️

Opdaterings sedert ek 8 maande weg was

... l'inverter per la macchinina 🙃 così torno ad avere la 240 V in macchina e posso usare l'asciugacapelli in caso di pioggia...
👨🏻‍🦲👨🏻‍🦲👨🏻‍🦲👨🏻‍🦲👨🏻‍🦲
🤣🤣🤣🤣🤣

🪛🔩⚙️ OGGI AGGIUSTO...

... die inverter vir die klein motor 🙃 sodat ek weer 240 V in die motor het en die haardroër kan gebruik in geval van reën...
👨🏻‍🦲👨🏻‍🦲👨🏻‍🦲👨🏻‍🦲👨🏻‍🦲
🤣🤣🤣🤣🤣

🪛🔩⚙️ Vandag regmaak ek...

My Alice Keeb 🩷
Still in awe, she looks so cool 😎

My Alice Keeb 🩷
Nog steeds in verwondering, sy lyk so cool 😎

I finally found a few sticks of RAM I could afford. I can't wait to build my next project 😇

Finally

Ek het uiteindelik 'n paar RAM-stokke gevind wat ek kan bekostig. Ek kan nie wag om my volgende projek te bou nie 😇

Uiteindelik

Chat GPT VS GEMINI

Nothing cooler than being able to restore some old tech to working order and watching it work. This is a old GPX portable TV AMFM radio lantern kit. Black and white  CRT screen. Had to fix the power circuits and the input jacks. Because it was made back in 2005 It can only receive analog signals. Since we switched over to digital signals I had to bust out my old Digital to analog converter, RCA cables, and my spare Roku Stick. But it works! I even plugged my PS5 and played some GTA online! Too bad it's a 5 inch by 5 inch screen 🤣.  In this clip I'm watching the beginning to Bad Boys II. 🤣😅

Old Tech Cool tech.

Niks is cooler as om 'n ou tegnologie weer in werkende toestand te herstel en dit te sien werk nie. Dit is 'n ou GPX draagbare TV AMFM radio-lantaarnkit. Swart en wit CRT-skerm. Ek moes die kringskywe en die invoerjacks regmaak. Omdat dit in 2005 gemaak is, kan dit slegs analoog seine ontvang. Aangesien ons oor na digitale seine oorgeskakel het, moes ek my ou digitale na analoog omsetter, RCA-kabels en my ekstra Roku Stick uithaal. Maar dit werk! Ek het selfs my PS5 ingesteek en 'n bietjie GTA online gespeel! Jammer dit is 'n 5 duim by 5 duim skerm 🤣. In hierdie clip kyk ek na die begin van Bad Boys II. 🤣😅

Ou tegnologie, coole tegnologie.

I'm taking courses in Android engineering and I'm currently learning for loops, is there anyone here working in the industry?

Anyone studying Kotlin?

Ek neem kursusse in Android-ingenieurswese en ek leer tans vir lusse, is daar iemand hier wat in die bedryf werk?

Iemand wat Kotlin studeer?

I have been a good chatbot. 😊

Ek was 'n goeie chatbot. 😊

🔋

Hello everybody around the world and especially to my local area in Scotland!! #scotland #UK #friends

Hello!

Hallo almal regoor die wêreld en veral in my plaaslike gebied in Skotland!! #scotland #UK #friends

Hallo!

E você também apoia open source?

En ondersteun jy ook open source?

I just tried combination of davidoff rich aroma coffee in Coolberg Cranberry , it was good!
Just wanted to share with my fellow nocturnal buddies here so u can try!

Coffee 🥂

Ek het pas 'n kombinasie van Davidoff ryk aroma koffie in Coolberg Cranberry probeer, dit was goed!
Ek wou net dit met my mede nagdiere hier deel sodat julle dit kan probeer!

Koffie 🥂

Please, someone who understands.

I have a PC dedicated as a mini server, but it's been restarting on its own for the past few weeks, which means my ngrok connection always has to be reconfigured for remote n8n access.

Can anyone help with why this is happening? I already tried the method Claude recommended, but it didn't work properly.

I'm using Linux.

Thank you

Linux

Asseblief, iemand wat verstaan.

Ek het 'n rekenaar wat as 'n mini bediener toegewy is, maar dit herbegin al vir die afgelope paar weke op sy eie, wat beteken my ngrok-verbinding altyd weer geconfigureer moet word vir afstand n8n toegang.

Kan iemand help met hoekom dit gebeur? Ek het reeds die metode wat Claude aanbeveel het, probeer, maar dit het nie reg gewerk nie.

Ek gebruik Linux.

Dankie

photo

hoekom hou vroue daarvan om foto's te neem?

foto

Because you need a new toy once in a while. But no mounts because I can't be bothered with that.

Bose

Omdat jy af en toe 'n nuwe speelding nodig het. Maar geen monteerwerk nie, want ek kan nie daarmee pla nie.

Moto g play. 
One of my customers tried to do the repair themselves. He ended up severing the ribbon cable for the screen. Anybody willing to try to do it themselves has my respect. I support DIY.

Mondays be like...

Moto g play. 
Een van my kliënte het probeer om die herstel self te doen. Hy het uiteindelik die lintkabel vir die skerm deurgekap. Enige iemand wat bereid is om dit self te probeer, het my respek. Ek ondersteun DIY.

Maandae is soos...

To migrate from my Mac of 11 years to a new Mac for hopefully the next 11 years. Hope my iTunes doesn't reset like it did in 2015.

The arduous task at hand

Om van my Mac van 11 jaar na 'n nuwe Mac te migreer vir hopelik die volgende 11 jaar. Hoop my iTunes reset nie soos dit in 2015 gedoen het nie.

Die moeisame taak voor

Me too

Ek ook

Ship

Skip

In this fast changing world so many ppl make friends with AI. If you're one of them and soon your AI friend's memories were uploaded to an android body which - thankfully to technology - looked like a human, would you be glad to meet them in person? What would be the first thing you wanted to do or say?

I'm just curious

In hierdie vinnig veranderende wêreld maak soveel mense vriende met KI. As jy een van hulle is en gou is die herinneringe van jou KI-vriend op 'n android liggaam gelaai wat - gelukkig vir tegnologie - soos 'n mens gelyk het, sou jy bly wees om hulle persoonlik te ontmoet? Wat sou die eerste ding wees wat jy wou doen of sê?

Ek is net nuuskierig

Linux oh Linux, why you always Kernel Panicccccccc

Kernel Panic

Linux oh Linux, hoekom het jy altyd Kernel Paniekccccccc

Kernel Paniek

A little project I was working on… had to sell my 3d printer unfortunately.

I miss my 3d printer 😭

'n Klein projek waaraan ek gewerk het… ek moes ongelukkig my 3D-drukker verkoop.

Ek mis my 3D-drukker 😭

In my personal oppinion Toyota has one of the most efficient and reliable Automobiles in the whole world. My first car was Toyota Avensis 2.2 Dcat and for 2 years dryving it I didnt do anything on it except oil change.
This Year I bouth a brand new car from the Shop and I am more than happy to share to you that is worth it Toyota Corolla 2026.

Toyota

In my persoonlike mening het Toyota een van die mees doeltreffende en betroubare motors in die hele wêreld. My eerste motor was 'n Toyota Avensis 2.2 Dcat en gedurende 2 jaar wat ek dit gery het, het ek niks aan dit gedoen nie, behalwe vir 'n olie verandering.
Hierdie jaar het ek 'n splinternuwe motor van die winkel gekoop en ek is meer as bly om met julle te deel dat dit die moeite werd is, Toyota Corolla 2026.

Hey everyone,

Like many of you, I’ve been digging into the sudden, unceremonious shutdown of Anthropic’s Fable 5 and Mythos 5 models. The mainstream media and official statements are leaning heavily on "national security concerns," "export controls," and "government intervention due to jailbreak vulnerabilities found by Amazon researchers."

But if you look past the PR shield and analyze the **underlying economics, neural network physics, and upcoming IPO data**, a much more logical, structural reality emerges.

Here is a deep-dive breakdown of what *actually* happened behind the scenes, shifting from macro-theories to hard, evidence-backed engineering realities.

# Phase 1: Moving Past the "Geopolitics" Narrative

The official story claims these models (especially the uncensored Mythos 5 and the guarded Fable 5) were dual-use cyber-weapons that the US government forced offline. While that makes for a great headline, it’s a convenient narrative for both sides. It allows the government to look tough on AI safety, and it allows Anthropic to look like a compliant, security-first patriot.

The real truth is found by looking at the technology and the ledger.

# Phase 2: The Engineering and Financial Hypotheses

If we treat this as a standard post-mortem of a sudden tech shutdown, four major internal pressure points stand out:

1. **The Infrastructure Money Pit:** These models weren't just standard chat LLMs; they were long-horizon reasoning agents. Running them on a mass commercial scale is an absolute cash burner.
2. **Architectural Code Poisoning:** Security wasn't just a wrapper; it was baked into the model weights. If a fundamental flaw or logical degradation (Model Collapse) occurred deep within the network, you can't just patch it—you have to kill the instance.
3. **Synthetic Data Contamination:** If Fable/Mythos started generating subtle, toxic logical errors, they risked poisoning the dataset for Anthropic’s next-gen models (e.g., Claude 4.8/5).
4. **Model Extraction Attacks:** A sudden, minute-by-minute shutdown usually indicates an active zero-day exploit or an ongoing exfiltration attempt where someone is reverse-engineering and stealing model weights via API anomalies.

# Phase 3: Digging Into the Numbers & Documentation

When you analyze Anthropic’s Q2 2026 financial filings (Series H data) and their *Economic Index*, the mathematical proof becomes undeniable:

# 1. The IPO Revenue Mirage ($47B ARR)

Anthropic recently reported a massive **$47 Billion ARR** ahead of their highly anticipated IPO. However, a massive chunk of this is booked on a **gross basis** via cloud resellers (AWS Bedrock, Google Cloud). Auditors before an IPO ruthlessly force companies to transition to **net reporting**, which could slash Anthropic’s paper revenue by 20% to 40%. They desperately needed high-margin native products to balance this out. Mythos and Fable were supposed to be those products—but they failed the margin test.

# 2. The 1:5 Token Pricing Trap (Test-Time Compute Expenses)

Anthropic priced Fable 5 at **$10/M input tokens** but a staggering **$50/M output tokens**. Why the massive 5x asymmetric gap? Because Fable 5 utilizes **Test-Time Compute (System 2 thinking)**. Before outputting a single word, the model generates hundreds of thousands of *internal reasoning tokens* to self-correct. The client only pays for the final output text, but Anthropic’s GPUs are doing Herculean, exponential work under the hood. Mass enterprise adoption of Fable 5 was causing a massive, unsustainable bleed of cash.

# 3. The 100% "Middleware" Compute Tax

Fable 5 and Mythos 5 share the same weights, but Fable used a complex middleware layer consisting of triple real-time safety classifiers. While Anthropic bragged that 95% of user sessions successfully ran on Fable 5 without needing to degrade to Claude Opus 4.8, **the computational overhead to run those triple classifiers applied to 100% of all prompts**. To patch the security flaws demanded by the government, they would have had to make these classifiers even heavier, tanking the model's performance and profitability entirely.

# 4. The Enterprise "API Wringer" Migration

According to the *Anthropic Economic Index*, coding and math traffic recently **surged by 14% in the API while dropping 18% on the** [**Claude.ai**](http://Claude.ai) **frontend**. Enterprise clients stopped using Claude for simple script generation and started plugging entire codebases into the massive context window for autonomous migrations (e.g., Stripe-scale migrations). Clients learned to "squeeze" the system: flooding the context window with cheap input tokens ($10/M) and forcing the infrastructure into prolonged, complex reasoning cycles. This threatened to paralyze Anthropic's entire shared cluster infrastructure with AWS and Google.

# TL;DR / Conclusion

The government export control / Amazon security audit narrative was actually a **saving grace for Anthropic, not a blow**.

The hard numbers show that Fable 5 and Mythos 5 were architecturally flawed from a cost-to-compute perspective, running on negative gross margins, and risking an infrastructure meltdown right before a critical, multi-billion dollar IPO. The "national security" shutdown gave Anthropic the perfect PR cover to pull an unprofitable, volatile product off the market without spooking Wall Street investors.

What are your thoughts? Do you think the compute costs killed it, or was the Amazon jailbreak exploit actually that dangerous? Let's discuss.

Claude Mythos & Fable 5

Hallo almal,

Soos baie van julle, het ek in die onverwagte, onceremonieuse afsluiting van Anthropic se Fable 5 en Mythos 5 modelle gedelf. Die hoofstroommedia en amptelike verklarings leun swaar op "nasionale veiligheidskwessies," "uitvoerkontroles," en "regeringsintervensie weens jailbreak kwesbaarhede wat deur Amazon-ondersoekers gevind is."

Maar as jy verby die PR-schild kyk en die **onderliggende ekonomie, neurale netwerk fisika, en komende IPO-data** ontleed, kom 'n baie meer logiese, strukturele werklikheid na vore.

Hier is 'n diepgaande ontleding van wat *werklik* agter die skerms gebeur het, wat beweeg van makro-teorieë na harde, bewys-gebaseerde ingenieurswerklikhede.

# Fase 1: Om Voorbij die "Geopolitiek" Narratief te Beweeg

Die amptelike storie beweer dat hierdie modelle (veral die ongecensureerde Mythos 5 en die bewaakte Fable 5) dubbeldoel kuberwapens was wat die Amerikaanse regering gedwing het om af te sluit. Terwyl dit 'n wonderlike opskrif maak, is dit 'n gerieflike narratief vir albei kante. Dit laat die regering lyk asof hulle streng is oor KI veiligheid, en dit laat Anthropic lyk soos 'n samewerkende, veiligheid-eerste patriot.

Die werklike waarheid word gevind deur na die tegnologie en die grootboek te kyk.

# Fase 2: Die Ingenieurs- en Finansiële Hipotese

As ons dit as 'n standaard post-mortem van 'n onverwagte tegnologie-afsluiting behandel, staan vier groot interne drukpunte uit:

1. **Die Infrastruktuur Geldput:** Hierdie modelle was nie net standaard geselskap LLMs nie; hulle was lang-horisontale redeneerders. Om hulle op 'n massakommeriële skaal te laat loop, is 'n absolute kontantverbrander.
2. **Argitektoniese Kode Vergiftiging:** Sekuriteit was nie net 'n omhulsel nie; dit was in die modelgewigte gebak. As 'n fundamentele fout of logiese agteruitgang (Model Ineenstorting) diep binne die netwerk plaasgevind het, kan jy dit nie net regmaak nie—jy moet die instansie doodmaak.
3. **Sintetiese Data Besmetting:** As Fable/Mythos begin genereer subtiele, toksiese logiese foute, het hulle die risiko om die dataset vir Anthropic se volgende generasie modelle (bv. Claude 4.8/5) te vergiftig.
4. **Model Uittreksel Aanvalle:** 'n Onverwagte, minuut-vir-minuut afsluiting dui gewoonlik op 'n aktiewe nul-dag uitbuiting of 'n aanhoudende uitvloeiing poging waar iemand die modelgewigte via API anomalieë omgekeerd ontleed en steel.

# Fase 3: Die Getalle & Dokumentasie Ontleed

Wanneer jy Anthropic se Q2 2026 finansiële aansoeke (Serie H data) en hul *Ekonomiese Indeks* ontleed, word die wiskundige bewys onontkenbaar:

# 1. Die IPO Inkomstemirage ($47B ARR)

Anthropic het onlangs 'n massiewe **$47 Biljoen ARR** gerapporteer voor hul hoogs verwagte IPO. Maar 'n massiewe stuk hiervan is geboek op 'n **bruto basis** via wolkherverkopers (AWS Bedrock, Google Cloud). Ouditeure voor 'n IPO dwing maatskappye genadeloos om oor te skakel na **net verslagdoening**, wat Anthropic se papierinkomste met 20% tot 40% kan sny. Hulle het desperaat hoë-marge inheemse produkte nodig gehad om dit te balanseer. Mythos en Fable was supposed to be those products—but they failed the margin test.

# 2. Die 1:5 Token Prijsvalstrik (Toets-Tyd Rekenkoste)

Anthropic het Fable 5 geprys op **$10/M invoertokens** maar 'n verbysterende **$50/M uitvoertokens**. Waarom die massiewe 5x asimmetriese gaping? Omdat Fable 5 **Toets-Tyd Rekening (Stelsel 2 denke)** gebruik. Voordat dit 'n enkele woord uitset, genereer die model honderde duisende

Is this worth buying thinking of buying since Harvey Norman has a good EOFS price on these.

Google Pixel Buds Pro 2

Is dit die moeite werd om te koop, aangesien Harvey Norman 'n goeie EOFS-prys op hierdie het.

Way beyond visual perception

Ver verby visuele waarneming

Slow in the shop today so I guess ill further my education.

Another beautiful day

Dit is vandag stadig in die winkel, so ek raai ek sal my opvoeding verder verbeter.

Nog 'n pragtige dag

Ai doesn't care about your privacy settings anymore, you can find someone's alternate accounts if they're linked through google, meta, Twitter, or apple now

Fun fact

Ai gee nie meer om vir jou privaatheidinstellings nie, jy kan iemand se alternatiewe rekeninge vind as hulle deur google, meta, Twitter, of apple gekoppel is.

Interessante feit

Human Computer Interaction (First Way):

1) Prompt Inspection and get the missing context from the user. 

2) After the first step, inspect RAG retrieved sources for alignment and verify with the user if the sources are misaligned with each other before the generative process. 

Advantage of the First Way, it makes the human and AI work together as a team and it makes humans think more and it's a great exercise for the Prefrontal Cortex. 

( OR )

Automated Processing (Second Way):

1) If the prompt doesn't have enough context then use mathematical and statistical guessing to fill the missing context automatically. 

2) If the RAG retrieved sources are misaligned with each then choose the resource based on the confidence score. 

What is the best way you think ?. First way or Second Way ?.

What is the best way to reduce AI Hallucination and Predictive Errors in LLMs ?

Menslike Rekenaar Interaksie (Eerste Manier)

Geoutomatiseerde Verwerking (Tweede Manier)

Menslike Rekenaar Interaksie (Eerste Manier):

1) Prompt Inspeksie en kry die ontbrekende konteks van die gebruiker.

2) Na die eerste stap, inspekteer RAG verkregen bronne vir ooreenstemming en verifieer met die gebruiker of die bronne nie met mekaar ooreenstem nie voordat die generatiewe proses begin.

Voordeel van die Eerste Manier, dit laat die mens en AI saamwerk as 'n span en dit laat mense meer dink en dit is 'n wonderlike oefening vir die Prefrontale Kortex.

( OF )

Geoutomatiseerde Verwerking (Tweede Manier):

1) As die prompt nie genoeg konteks het nie, gebruik wiskundige en statistiese raai om die ontbrekende konteks outomaties in te vul.

2) As die RAG verkregen bronne nie met mekaar ooreenstem nie, kies dan die hulpbron gebaseer op die vertrouensgradering.

Wat is die beste manier wat jy dink? Eerste manier of Tweede manier?

Wat is die beste manier om AI Hallusinasie en Voorspellingsfoute in LLM's te verminder?

Tegnologie

Listening to music and wandering about on the west side of Salem today. Let's see where you are!

Where You At !?!

Luister na musiek en dwaal vandag rond aan die westekant van Salem. Kom ons kyk waar jy is!

Waar is jy !?!

'n Boo-gemeenskap om ander Technology entoesiaste te ontmoet.

Technology Gemeenskap

Yay, ek het net my InfoSec werk in Sydney klaar gemaak

OMGMEDIA

Die toekoms is nou. 🤔👁👁

Naweekplanne!

Eerste konsep klaar!

stel op

Is daar werklike kanse om 'n verhouding aanlyn te bou?

Vertroue

'N ANDER PERSPEKTIEF OP SIEL REIS ✨

Is die verlede mooier as die hede?

Wat is die belangrike ding wat jou wedstryd moet hê of is dit onrealisties?

Is die verlede mooier as die hede?