Software-Developer, Linux/Debian Server-Administrator, GalacticShift I do NOT give any financial advice at any time!
Joined July 2022
- Tweets4.5K
- Following120
- Followers6.8K
- Likes2.8K
StrathCole retweeted
Hello #TerraClassic $lunc.
@igorsoares62 followed the PPJ playbook like a pro.
The delay on the task is nothing of his wrongdoing but simply due to the inertia of his peers he had to work with to get the validator network running.
@igorsoares62 delivered. Pay him!
Or what signal are we sending to other builders? We request them to work under PPJ and then we are not even able to stick to our own rules and make the payments under our own terms?
Yes... this is the right kind of instability signal that every $lunc hater has been expecting from us since the start. Secrectly, they rub their hands.
With this behavior we are now officially degraded to a bunch of uncooperative whiny children.
Gosh, this makes me so angry.
StrathCole retweeted
Good to see the website getting better by many contributors helping out
Lunc = Community
terra-classic.io/
Everyone is free to help, update, contribute
How awesome is that ?!
StrathCole retweeted
Hey! 😍
I just released version v4.0.2 of the terrad client:
This time it's a rolling release WITH NO COORDINATED UPGRADE NEEDED. Validators and operators can install the new client at their disposal.
Whats new?
- Incorporates the disclosed wasm security fix by @cosmoslabs_io and bumps the affected dependencies to the official released versions (instead of pointing to the private repositories)
- featuring a new in-place-testnet command to streamline future upgrade testing and validating upgrades against live mainnet state.
- some important cleanup items by a new contributor (GeoffreySHD) improving dyncomm genesis roundtrip, removing older debug prints from @orbit__labs.
Thanks to @ColeStrathclyde for the assistance.
#blamefrag
StrathCole retweeted
Around $150 of $USDC (Noble minted) community liquidity just crossed 4 blockchains and came home to #TerraClassic as $USDC (Injective minted) into the 12171 Liqudity Initiatives DAO wallet.
Prop 12226 rehearsal: ✅ DONE 🧵👇
---
What happened:
Terra Classic → Noble → Ethereum → Injective → back to Terra Classic.
149.926928 $USDC.n left the "Terra Classic Liquidity Initiative 12171" multisig and 149.926478 $USDC.inj arrived back.
⏱ About 35 minutes of bridging from execution to arrival
🔐 Every step on the multisig was approved on chain by the DAO (7 of 10 signers needed)
🔵 Native $USDC burned and minted by Circle's CCTP on both bridge hops
Prop 12226 moves the terra-luna:native / $USDC liquidity from Noble $USDC to Injective $USDC. Before touching the main liquidity, we withdrew about $100 from each of the 3 pools (@_Terraport_, @garuda_defi, @terraswap_io) and ran the full route end to end.
The route works. Every tx can be verified on chain. We prepared a discourse thread for this to keep you posted.
Next:
1️⃣ Set up the new terra-luna:native / $USDC (Injective) pools with the rehearsal funds
2️⃣ Then the full migration along the same route
Full report with every tx link - see answer to this post.
And as always: #blamefrag
#TerraClassic terra-luna:native $USTC $USDC
StrathCole retweeted
Currently preparing a terrad in-place-testnet command for the #TerraClassic node client.
I have seen something like this on @osmosis and also @ZIGChain and other major networks within the Cosmos ecosystem.
Let's have a closer look.
1/x
StrathCole retweeted
📣📣📣
There is an important and mandatory update to be applied to Terra Classic.
discourse.luncgoblins.com/t/…
I won't leave the discussion up for seven days. On-chain prop goes out sooner than later.
Thanks to @ColeStrathclyde & @VegasMorph for guidance on this difficult one.
StrathCole retweeted
If we do not act on this, USDC.n will get stuck at some point and the community owned on-chain LUNC/USDC.n liquidity positions amounting to ~$130k will become worthless.
So whoever votes NO on this one is showing his/her true colors.
Having @ChatGPT Astra build a local monitoring tool to its own token usage is fun 😅
Anyone seen the #easteregg in @ChatGPT image? No idea how long it has been there already 😂
Maybe I'll try something like this and see if it does any good. 😅
## Tool efficiency
Before research, identify likely information needs. Batch or parallelize independent, read-only lookups; use sequential calls for dependencies. Keep queries targeted and output bounded. Reuse results unless stale. Avoid speculative bulk retrieval. Minimize unnecessary round trips and context growth without sacrificing correctness or verification.
I think people underestimate the effect of context length and tool calls on token usage.
Imagine a current context length of 15,000 tokens. The model then researches your codebase by doing multiple tool calls like terminal execution or MCP. Let's say it needs 10 tool calls.
Not even considering the increase in context due to the calls themselves, this would be 150,000 input tokens (mostly cached).
Now let's say the same code research happens later in the session at a context length of maybe 150,000 tokens.
Boom, 1,500,000 input tokens (mostly cached).
Depending on how many tool calls the model needs in what stage of the session, token usage can differ greatly. And imho this has a huge impact on how fast your token quota drains in @ChatGPT or @claudeai or whatever other provider you're using.
Each sequential model → tool → model round trip generally causes another model call over the increased context. Multiple parallel tool calls could share a single round trip.
And the output of tool calls (depending on the implementation etc.) normally reaches the next turn's input again.
I guess there's a lot of room for optimization here. Maybe someone has a clever idea.
Cheers.
I think people underestimate the effect of context length and tool calls on token usage.
Imagine a current context length of 15,000 tokens. The model then researches your codebase by doing multiple tool calls like terminal execution or MCP. Let's say it needs 10 tool calls.
Not even considering the increase in context due to the calls themselves, this would be 150,000 input tokens (mostly cached).
Now let's say the same code research happens later in the session at a context length of maybe 150,000 tokens.
Boom, 1,500,000 input tokens (mostly cached).
Depending on how many tool calls the model needs in what stage of the session, token usage can differ greatly. And imho this has a huge impact on how fast your token quota drains in @ChatGPT or @claudeai or whatever other provider you're using.
Each sequential model → tool → model round trip generally causes another model call over the increased context. Multiple parallel tool calls could share a single round trip.
And the output of tool calls (depending on the implementation etc.) normally reaches the next turn's input again.
I guess there's a lot of room for optimization here. Maybe someone has a clever idea.
Cheers.
So I had a look at all of them with a fresh mind now.
Here's my personal ranking, just from visuals.
#1 @ChatGPT Sol High
#2 @Alibaba_Qwen 3.8 Flash
#3 @Zai_org GLM 5.3 Flash
#4 GPT Astra High
#5 @claudeai Opus 5 High
#6 GPT Astra Low
#7 Qwen 3.8 27b
#8 @deepseek_ai V4 Flash
Considering the current pricing, price-result-winners are GLM Flash and Qwen Flash.
I wouldn't publish any of them 1:1, for sure do some iterations.
So with everybody doing comparisons, I though I give it a quick and simple shot, too with gpt-6-astra.
Simple prompt on an empty folder in codex cli.
"You are a senior webdesigner. Create the most beatiful sci-fi themed website you can imagine. NO REVIEWS. No subagents. NO skill usage."
Only tested GPT 5.6 Sol high, GPT 6 Astra low, GPT 6 Astra high, Claude Opus 5, GLM 5.3 Flash.
Sol first. It took 8m33s to complete. Token usage: 86,494 (+ 1,222,144 cached). (others in the comments)
After doing Terra xHigh and Luna Max, I would rank them somewhere ~#6-#7.
xitter.cf/ColeStrathclyde/status…
xitter.cf/ColeStrathclyde/status…
So with everybody doing comparisons, I though I give it a quick and simple shot, too with gpt-6-astra.
Simple prompt on an empty folder in codex cli.
"You are a senior webdesigner. Create the most beatiful sci-fi themed website you can imagine. NO REVIEWS. No subagents. NO skill usage."
Only tested GPT 5.6 Sol high, GPT 6 Astra low, GPT 6 Astra high, Claude Opus 5, GLM 5.3 Flash.
Sol first. It took 8m33s to complete. Token usage: 86,494 (+ 1,222,144 cached). (others in the comments)
Here's Qwen3.8 27b through the @Hetzner_Online experiments inference. Took a lot of time (the experiments api is quite slow). Cost: free.
