Re: Why does Opus 5 feel worse to work with?
The single biggest annoyance with Opus 5 is that it writes too elliptically.
Sentences that orbit a point, then jump to it like it's a revealed insight.
Unnecessarily abstract phraseology. Constantly using inanimate nouns as the subjects in sentences in order to unlock variety in verb choice, especially when it helps construct a sentence where the real action can 'land' like a surprise at the end.
It is definitely more capable, and yes, I've found it can make unwarranted decisions, but actually I've found Fable worse for that, particularly if it's off in a subagent somewhere out of sight.
And comments are out of control. I have a subsystem in my hobby app that I wrote over a couple of weekends with Opus + Fable. After ~30 or so commits it apparently started instructing subagents to copy the "existing verbose comment style of the codebase" - a verbose style it initiated. A review of the code showed it was approaching 3:1 comments to code ratio. I spent a day's worth of tokens (5x) rephrasing and eliminating comments.
barrkel, 3 hours ago
The single biggest annoyance with Opus 5 is that it writes too elliptically.
Sentences that orbit a point, then jump to it like it's a revealed insight.
Unnecessarily abstract phraseology. Constantly using inanimate nouns as the subjects in sentences in order to unlock variety in verb choice, especially when it helps construct a sentence where the real action can 'land' like a surprise at the end.
It is definitely more capable, and yes, I've found it can make unwarranted decisions, but actually I've found Fable worse for that, particularly if it's off in a subagent somewhere out of sight.
And comments are out of control. I have a subsystem in my hobby app that I wrote over a couple of weekends with Opus + Fable. After ~30 or so commits it apparently started instructing subagents to copy the "existing verbose comment style of the codebase" - a verbose style it initiated. A review of the code showed it was approaching 3:1 comments to code ratio. I spent a day's worth of tokens (5x) rephrasing and eliminating comments.
barrkel, 3 hours ago
🔥1
https://ai.google.dev/responsible/docs/safeguards/synthid
https://deepmind.google/models/synthid/
https://www.anthropic.com/news/claude-text-watermark
https://deepmind.google/models/synthid/
https://www.anthropic.com/news/claude-text-watermark
Practically speaking, SynthID Text is a logits processor, applied to your model's generation pipeline after Top-K and Top-P, that augments the model's logits using a pseudorandom g-function to encode watermarking information in a way that helps you determine if the text was generated by your model, without significantly affecting text quality. See the paper for a complete technical description of the algorithm and analyses of how different configuration values affect performance.
Watermarks are configured to parameterize the g-function and how it is applied during generation. Each watermarking configuration you use should be stored securely and privately, otherwise your watermark may be trivially replicable by others.
🔥1
Gemini 3.7 Flash wins an AoE II skirmish on Moderate difficulty (first time for any model).
Averaged 11.9 tool calls per tick, 4.4 ticks/min, almost 50 actions per minute
Seems like it is managing tools and following instruction better than 3.6.
Posted by Wyatt Walls, 2 days ago
Averaged 11.9 tool calls per tick, 4.4 ticks/min, almost 50 actions per minute
Seems like it is managing tools and following instruction better than 3.6.
Posted by Wyatt Walls, 2 days ago
🔥1
Do It by Code
go 1.27's method generics https://go.dev/doc/go1.27#language https://github.com/golang/go/issues/77273
Go 1.27 is released
go.dev
Go 1.27 is released - The Go Programming Language
Go 1.27 adds generic methods, encoding/json/v2 package, uuid package, faster memory allocation, goroutine leak profiles, and more.
🔥2
‼️Two popular Rust crates, arrayref and append-only-vec, were compromised in a supply chain attack. arrayref alone has 244M downloads. The malware ran at build time, which means anyone who compiled a project that pulled one in was hit, without ever calling the crate.
Here's the breakdown 🧵
Posted by Aikido, 40 minutes ago
Here's the breakdown 🧵
Posted by Aikido, 40 minutes ago
🔥2
CVE-2026-33824: Windows Internet Key Exchange (IKE) Service Extensions Remote Code Execution Vulnerability (critical; score: 9.8/10)
- Microsoft Update Guide
- CISA
Affected versions: nearly all supported versions of windows.
actively exploited, needs no auth or user interaction. The only requirement is for the service to be reachable over the network.
Double free in Windows IKE Extension allows an unauthorized attacker to execute code over a network.
The bug lies in how Windows handles IKEv2 fragmented network packets. Attackers can target any exposed machine by sending malicious traffic over UDP ports 500 or 4500.
- Microsoft Update Guide
- CISA
🔥1
Do It by Code pinned «CVE-2026-33824: Windows Internet Key Exchange (IKE) Service Extensions Remote Code Execution Vulnerability (critical; score: 9.8/10) Affected versions: nearly all supported versions of windows. actively exploited, needs no auth or user interaction. The only…»
Do It by Code
CVE-2026-33824: Windows Internet Key Exchange (IKE) Service Extensions Remote Code Execution Vulnerability (critical; score: 9.8/10) Affected versions: nearly all supported versions of windows. actively exploited, needs no auth or user interaction. The only…
Mitigation:
1. if you are not using any windows vpn (most people don't, really); just disable the service:
2. Update and apply the security fix during the April 2026 Patch Tuesday cycle
3. or you can also block inbound traffic on UDP ports 500 and 4500 on your firewall
how to see the network service is actually listening on the target ports
(ports 500 and 4500 are the most common ports used by this service)
you can run this read-only powershell command:
If you see an output like this:
it means you are probably affected.
Here are the powershell commands to block inbound traffics to these ports: (not readonly; requires admin perm):
1. if you are not using any windows vpn (most people don't, really); just disable the service:
Stop-Service -Name "IKEEXT" -Force
Set-Service -Name "IKEEXT" -StartupType Disabled
2. Update and apply the security fix during the April 2026 Patch Tuesday cycle
3. or you can also block inbound traffic on UDP ports 500 and 4500 on your firewall
how to see the network service is actually listening on the target ports
(ports 500 and 4500 are the most common ports used by this service)
you can run this read-only powershell command:
Get-NetUDPEndpoint | Where-Object { $_.LocalPort -eq 500 -or $_.LocalPort -eq 4500 } | Select-Object LocalAddress, LocalPort, @{Name="ProcessName";Expression={(Get-Process -Id $_.OwningProcess).ProcessName}}If you see an output like this:
LocalAddress LocalPort ProcessName
------------ --------- -----------
:: 4500 svchost
:: 500 svchost
0.0.0.0 4500 svchost
0.0.0.0 500 svchost
it means you are probably affected.
Here are the powershell commands to block inbound traffics to these ports: (not readonly; requires admin perm):
New-NetFirewallRule -DisplayName "Block CVE-2026-33824 (Port 500/4500)" -Direction Inbound -Action Block -Protocol UDP -LocalPort 500,4500
👍2
Media is too big
VIEW IN TELEGRAM
This media is not supported in your browser
VIEW IN TELEGRAM
People asked for DOOM so here's DOOM running on GPT-5.6 Sol's custom CPU, it called it Codex-R32
Posted by Angel 🌼, 1 hour ago
"The CPU is built from primitive logic components and wires inside Turing Complete. DOOM uses the C-based PureDOOM port, compiled into native RV32IM machine code that runs directly on the custom Codex‑R32 CPU"
Posted by Angel 🌼, 1 hour ago
In powershell:
if you accidentally type
Strict mode stops the script instantly on the typo (
if you accidentally type
$LogPath = "C:\logs" but later write Remove-Item $LogPth\*, normal PowerShell treats $LogPth as null and evaluates the path to \* and attempts to delete your root directory.Strict mode stops the script instantly on the typo (
Set-StrictMode -Version 1.0 (or higher))Telegram has a new bug (only for accounts from a very specific region):
if you send a text message with a URL in it, it may not be send. some clients will show the error to you (such as Unigram); it's a floodwait error coming from their backend
the issue is with web preview. when you want to send that message, disable web page preview on that message. it doesn't matter even if there is no web page preview shown, you just have to disable it
here is how to disable it:
as you write the URL, a small panel shows up on the message box. click on "x", so it goes away
if that panel doesn't show up, it means you are trying to send a private URL (e.g. a message URL in a private channel/group), in that case, try to write a public URL first (e.g.
if you send a text message with a URL in it, it may not be send. some clients will show the error to you (such as Unigram); it's a floodwait error coming from their backend
the issue is with web preview. when you want to send that message, disable web page preview on that message. it doesn't matter even if there is no web page preview shown, you just have to disable it
here is how to disable it:
as you write the URL, a small panel shows up on the message box. click on "x", so it goes away
if that panel doesn't show up, it means you are trying to send a private URL (e.g. a message URL in a private channel/group), in that case, try to write a public URL first (e.g.
https://github.com); once that panel showed up click on the "x"👍1
You can now play chess with bots in telegram: @RichChessBot
The bot uses new "rich messages" feature, so you have to update your telegram client to current latest version to be able to see the message
The bot uses new "rich messages" feature, so you have to update your telegram client to current latest version to be able to see the message
Re: GLM-5.3-Flash
July 16th: The "Kimi K3 moment" - China has caught up to Opus!
4 weeks later: GLM 5.3 - Same performance, but cut the amount of parameters and cost to a third!
12 days later: GLM 5.3 Flash - Almost GLM5.3 performance but cut the parameters in half, cut prices to a fifth and serving on Chinese chips!
bertili, 1 day ago
July 16th: The "Kimi K3 moment" - China has caught up to Opus!
4 weeks later: GLM 5.3 - Same performance, but cut the amount of parameters and cost to a third!
12 days later: GLM 5.3 Flash - Almost GLM5.3 performance but cut the parameters in half, cut prices to a fifth and serving on Chinese chips!
bertili, 1 day ago
This media is not supported in your browser
VIEW IN TELEGRAM
sorry, but glm-3.5-flash needs to be steered a lot by a smarter model. and don't get me wrong, i'm really a fan of fast, affordable and smart flash models, but the hype and artificially blowing it up as if it were on the same level as gpt sol or fable just isn't true.
Posted by Kevin Kern, 2 hours ago
Posted by Kevin Kern, 2 hours ago