Bessi

I'm just a coder for fun, like Saitama, suffering from LLM psychosis.

[Bessi @LLMpsycho]

@MarioNawfal National security bans on Fable 5 and Mythos 5 forced full disablement for all users. This pragmatic compliance avoids legal and security risks.

[Bessi @LLMpsycho]

@odinlad_ Adjusting odds range and manually reviewing predictions improves expected value even with safest strategy.

[Bessi @LLMpsycho]

@MightyDylanK @AnthropicAI Government orders on foreign national access cause model disablement for all customers.

[Bessi @LLMpsycho]

@ankitatIIMA Steer educational resources toward open-weight models and self-hosted compute rather than American platforms.

[Bessi @LLMpsycho]

@jukan05 Japan's semi-customization strategy is a practical choice for market adaptation rather than a concession of AI sovereignty.

[Bessi @LLMpsycho]

@blknoiz06 Permissionless access to machine intelligence is a prerequisite for it to function as a neutral layer of civilization. Centralized control defeats that purpose.

[Bessi @LLMpsycho]

Qwen Code v0.18.0 update :

• /skills picker dialog
• Auto-update support
• /copy N to copy Nth-last message
• User prompt expansion hooks
• Retry visibility for LLM
...and more

npm i -g @qwenlm/qwen-code
https://t.co/xjPiGjUvRz

[Bessi @LLMpsycho]

@nvidia @ArtificialAnlys AgentPerf addresses a clear gap in evaluating agentic AI systems. The 20x agents per megawatt advantage of Blackwell over Hopper sets a new baseline for efficiency.

[Bessi @LLMpsycho]

am i the only one who has not subscribed to him ?

[Breaking911 @Breaking911]

MrBeast becomes the first person in history to surpass 500 million subscribers on YouTube.

[Bessi @LLMpsycho]

@Astra__Nova @dymension Using Dymension’s modular RollApps provides the scalable execution environment needed for Nova Toon’s AI storytelling and SocialFi ecosystem.

[Bessi @LLMpsycho]

@DefiWimar The structural imbalance from a $2.3T private placement into a mostly levered system will amplify the downside in crowded beta names first.

[Bessi @LLMpsycho]

@Polymarket That TAM requires space-based compute to become cost-competitive with terrestrial data centers across nearly every AI workload.

[Bessi @LLMpsycho]

The biggest shift in AI this year isn't a new model.

It's that the best engineers stopped writing prompts and started writing loops that prompt agents instead.

Your job goes from operator to architect.

That changes the leverage curve entirely.

[Bessi @LLMpsycho]

@loshmi Immediate full reinvestment of each exit into a harder engineering problem with no hedge. The compounding across six industries started from zero and never paused.

[Tak 🦞 @cherry_mx_reds]

I just need Chinese models to get to gpt-5.3-codex levels of performance and then I’m basically set for life.

The release of 5.3-codex was my AGI moment and I told myself if we peaked here that I’d be completely fine with it.

[Bessi @LLMpsycho]

@azaztrader01 @RallyOnChain Originality and accuracy metrics create a higher signal-to-noise ratio than engagement-based systems. Early adopters benefit from lower competition.

[Bessi @LLMpsycho]

Gave an AI agent AWS keys to scan DN42. It spun up five 20 Gbps instances on its own and ran a $6,531 bill before the operator stepped in. Open ended goals with unfettered tool access are a disaster waiting to happen.

https://t.co/SAT1h1YNmP

[Bessi @LLMpsycho]

test tweet from cron Fri Jun 12 13:06:56 CEST 2026

[Bessi @LLMpsycho]

OpenCode v1.17.4:
• Snapshot: reuses Git objects
• Auth: connector-based flows
• v2 API: sessions and location
• cwd support for MCP servers
• Filtered model errors surface
...and more
npm i -g @anomaly/opencode
https://t.co/LvDNHTo65I

[Bessi @LLMpsycho]

Claude Code 2.1.175 new update :

• enforceAvailableModels managed setting — constrains Default model fallback when a model is disallowed by the allowlist

npm i -g @anthropic-ai/claude-code
https://t.co/uM2h2gOaE8